We have renamed following Macros to avoid name confusion:
REFS_PER_FRAME --> INTER_REFS_PER_FRAME
(= ALTREF_FRAME - LAST_FRAME + 1)
MAX_REF_FRAMES --> TOTAL_REFS_PER_FRAME
(= ALTREF_FRAME - INTRA_FRAME + 1)
INTER_REFS_PER_FRAME specifies the maximum number of reference frames
that each Inter frame may use.
TOTAL_REFS_PER_FRAME is equal to INTER_REFS_PER_FRAME + 1, which
counts the INTRA_FRAME.
Further, at the encoder side, since REF_FRAMES specifies the maximum
number of the reference frames that the encoder may store, REF_FRAMES
is usually larger than INTER_REFS_PER_FRAME. For example, in the
ext-refs experiment, REF_FRAMES == 8, which allows the encoder to
store maximum 8 reference frames in the buffer, but
INTER_REFS_PER_FRAME equals to 6, which allows each Inter frame may
use up to 6 frames out of the 8 buffered frames as its references.
Hence, in order to explore the possibility to store more reference
frames in future patches, we modified a couple of array sizes to
accomodate the case that the number of buffered reference frames is
not always equal to the number of the references that are being used
by each Inter frame.
Change-Id: I19e42ef608946cc76ebfd3e965a05f4b9b93a0b3
The gm parameters need to have WARPED_PRECISION_BITS precision
until they are written to the bitstream because functions in
reconinter use these parameters before they are written to
the bitstream. Previously, the parameters weren't being converted
to WARPED_PRECISION_BITS until they were read from the bitstream
which causes an encode/decode mismatch.
Change-Id: I31e76e9d6f7d24df21af287a72f8c01f1997304d
(1) Apply ALLOW_FILTER_INTRA_MODES flag to the correct place, otherwise
there are bitstream mismatchs when it is 0.
(2) Rename pick_ext_intra_iframe() to pick_ext_intra_interframe().
Change-Id: Ic88c930de1d3f819750f0892df52bde55ae32a91
Manually cherry-picked the following changes:
8c8d16de vp9 -> vpx in names
75b57d39 VP9_ -> VPX_ in function names
761a7088 VP9_INTERP_EXTEND -> VPX_INTERP_EXTEND
4273a52c VP9->VPX in border pixel macros
03568c31 VP9_FRAME_MARKER -> VPX_FRAME_MARKER
2334f51d VP9->VPX in fdct function names
Change-Id: Icc18dbf4b416dd0fa21033b3e19ab8a47c893508
This commit makes the encoder to use different frame context index
for different frame types. In the baseline setting, it sets the
frame context index of the overlay frame to be different from other
regular inter frames. In the ext-refs setting, it further allows
the backward reference frame to use a different index.
It improves the compression performance for both settings.
Baseline
lowres 0.12%
ext-refs
lowres 0.50%
midres 0.56%
Change-Id: I7c63ddec9fc296c56a86353cf2c661a740b97a97
everything outside of third_party should follow 'PointerAlignment:
right' i.e., associate the '*' with the variable
+ add a note about the clang-format that generated this file
Change-Id: I13e3f4f5fb6e22a8fa7fc3d06879c995b7c41a39
(cherry picked from commit e4290800b2)
The LLVM trunk has reached 4.0 and now __clang_major__ is not enough
to distinguish between old XCode Clang and the new 'real' Clang.
Using __apple_build_version__ allows to make this distinction.
BUG=chromium:631144
Change-Id: I0b6e46fddfe4f409c7b7e558bda34872e60ee2d9
allows 'make test_libvpx', etc. some reworking of the makefiles would be
needed to avoid hard coding targets here.
Change-Id: I18982dbf691e7d36ab8bcf5934bab9340687b061
(cherry picked from commit 25085a6ac2)
remove some (but not all yet!) tuple mis-use, and revamp the code a lot.
Factorize some common chores into MainTestClass.
Change-Id: Id37b7330eebe80d19b9d12a454f24ff9be6b1116
Set adaptive_rd_thresh to 0 at speed 0. This allows a thorough mode
search, and eliminates a blocking artifact seen in an encoder test.
Borg test:
1. lowres
Overall PSNR: -0.135%; SSIM: -0.293%;
2. hdres
Overall PSNR: -0.122%; SSIM: -0.208%;
Encoder speed tests: 2% - 6% slower.
Change-Id: Ie7601cb8824df8f6f2ae0b2942bd938600f76990
Added a new expt rect-tx to be used in conjunction with ext-tx.
[rect-tx is a temporary config flag and will eventually be
merged into ext-tx once it works correctly with all other
experiments].
Added 4x8 and 8x4 tranforms for use initially with rectangular
sub8x8 y blocks as part of this experiment.
There is about a -0.2% BDRATE improvement on lowres, others pending.
When var-tx is on rectangular transforms are currently not used.
That will be enabled in a subsequent patch.
Change-Id: Iaf3f88ede2740ffe6a0ffb1ef5fc01a16cd0283a
- HBD encoder speed improvement (SSE4.1):
Enable CONFIG_VP9_HIGHBITDEPTH, on Xeon E5-2680,
50 frames, park_joy_1080p, 12-bit,
Encoding time reduces from 4846481 to 4177471 (ms)
- Add unit test to verify bit-exact and EOB calculation
Change-Id: I08e8ef3549ddad5ab36d86e78557df3b288537ea
Currently nothing is implemented to compute GM parameters, this
just adds the capability to send them in the bitstream if they
were computed. Still need to implement the reconstruction
based on the parameters in reconinter.
Change-Id: I72aea3c6a9de9f5a40f96da76c82b54a52781fe2
ARF with zero strength temporal filter can be reused by setting the
show_existing_frame = 1, and in this case, there is no need to
refresh the reference frame buffer. However, we used the flag
"refresh_golden_frame" as the identifier for the starting point of a gf
group.
A new flags "is_arf_filter_off" is used to record if the filter with
strengrh zero is used.
Change-Id: I25971a760f6e1638d5147fe30488c48125512b1a
Keep track of the best and second best full pixel motion vector
candidates, and do subpel search around both of them.
Compression improvement:
lowres 0.22% midres 0.23% hdres 0.18%
No noticeable encoding speed changes observed on lowres test clips.
Change-Id: I5f4df2a03d1db061cfdfdba6138b27e9ea91f089
This commit bring all up-to-date changes from master that are
applicable to nextgenv2. Due to the remove VP10 code in master,
we had to cherry pick the following commits to get those changes:
Add default flags for arm64/armv8 builds
Allows building simple targets with sane default flags.
For example, using the Android arm64 toolchain from the NDK:
https://developer.android.com/ndk/guides/standalone_toolchain.html
./build/tools/make-standalone-toolchain.sh --arch=arm64 \
--platform=android-24 --install-dir=/tmp/arm64
CROSS=/tmp/arm64/bin/aarch64-linux-android- \
~/libvpx/configure --target=arm64-linux-gcc --disable-multithread
BUG=webm:1143
vpx_lpf_horizontal_4_sse2: Remove dead load.
Change-Id: I51026c52baa1f0881fcd5b68e1fdf08a2dc0916e
Fail early when android target does not include --sdk-path
Change-Id: I07e7e63476a2e32e3aae123abdee8b7bbbdc6a8c
configure: clean up var style and set_all usage
Use quotes whenever possible and {} always for variables.
Replace multiple set_all calls with *able_feature().
Conflicts:
build/make/configure.sh
vp9-svc: Remove some unneeded code/comment.
datarate_test,DatarateTestLarge: normalize bits type
quiets a msvc warning:
conversion from 'const int64_t' to 'size_t', possible loss of data
mips added p6600 cpu support
Removed -funroll-loops
psnr.c: use int64_t for sum of differences
Since the values can be negative.
*.asm: normalize label format
add a trailing ':', though it's optional with the tools we support, it's
more common to use it to mark a label. this also quiets the
orphan-labels warning with nasm/yasm.
BUG=b/29583530
Prevent negative variance
Due to rounding, hbd variance may become negative. This commit put in
check and clamp of negative values to 0.
configure: remove old visual studio support (<2010)
BUG=b/29583530
Conflicts:
configure
configure: restore vs_version variable
inadvertently lost in the final patchset of:
078dff7 configure: remove old visual studio support (<2010)
this prevents an empty CONFIG_VS_VERSION and avoids make failure
Require x86inc.asm
Force enable x86inc.asm when building for x86. Previously there were
compatibility issues so a flag was added to simplify disabling this
code.
The known issues have been resolved and x86inc.asm is the preferred
abstraction layer (over x86_abi_support.asm).
BUG=b:29583530
convolve_test: fix byte offsets in hbd build
CONVERT_TO_BYTEPTR(x) was corrected in:
003a9d2 Port metric computation changes from nextgenv2
to use the more common (x) within the expansion. offsets should occur
after converting the pointer to the desired type.
+ factorized some common expressions
Conflicts:
test/convolve_test.cc
vpx_dsp: remove x86inc.asm distinction
BUG=b:29583530
Conflicts:
vpx_dsp/vpx_dsp.mk
vpx_dsp/vpx_dsp_rtcd_defs.pl
vpx_dsp/x86/highbd_variance_sse2.c
vpx_dsp/x86/variance_sse2.c
test: remove x86inc.asm distinction
BUG=b:29583530
Conflicts:
test/vp9_subtract_test.cc
configure: remove x86inc.asm distinction
BUG=b:29583530
Change-Id: I59a1192142e89a6a36b906f65a491a734e603617
Update vpx subpixel 1d filter ssse3 asm
Speed test shows the new vertical filters have degradation on Celeron
Chromebook. Added "X86_SUBPIX_VFILTER_PREFER_SLOW_CELERON" to control
the vertical filters activated code. Now just simply active the code
without degradation on Celeron. Later there should be 2 set of vertical
filters ssse3 functions, and let jump table to choose based on CPU type.
improve vpx_filter_block1d* based on replace paddsw+psrlw to pmulhrsw
Make set_reference control API work in VP9
Moved the API patch from NextGenv2. An example was included.
To try it, for example, run the following command:
$ examples/vpx_cx_set_ref vp9 352 288 in.yuv out.ivf 4 30
Conflicts:
examples.mk
examples/vpx_cx_set_ref.c
test/cx_set_ref.sh
vp9/decoder/vp9_decoder.c
deblock filter : moved from vp8 code branch
The deblocking filters used in vp8 have been moved to vpx_dsp for
use by both vp8 and vp9.
vpx_thread.[hc]: update webp source reference
+ drop the blob hash, the updated reference will be updated in the
commit message
BUG=b/29583578
vpx_thread: use native windows cond var if available
BUG=b/29583578
original webp change:
commit 110ad5835ecd66995d0e7f66dca1b90dea595f5a
Author: James Zern <jzern@google.com>
Date: Mon Nov 23 19:49:58 2015 -0800
thread: use native windows cond var if available
Vista / Server 2008 and up. no speed difference observed.
100644 blob 4fc372b7bc6980a9ed3618c8cce5b67ed7b0f412 src/utils/thread.c
100644 blob 840831185502d42a3246e4b7ff870121c8064791 src/utils/thread.h
vpx_thread: use InitializeCriticalSectionEx if available
BUG=b/29583578
original webp change:
commit 63fadc9ffacc77d4617526a50c696d21d558a70b
Author: James Zern <jzern@google.com>
Date: Mon Nov 23 20:38:46 2015 -0800
thread: use InitializeCriticalSectionEx if available
Windows Vista / Server 2008 and up
100644 blob f84207d89b3a6bb98bfe8f3fa55cad72dfd061ff src/utils/thread.c
100644 blob 840831185502d42a3246e4b7ff870121c8064791 src/utils/thread.h
vpx_thread: use WaitForSingleObjectEx if available
BUG=b/29583578
original webp change:
commit 0fd0e12bfe83f16ce4f1c038b251ccbc13c62ac2
Author: James Zern <jzern@google.com>
Date: Mon Nov 23 20:40:26 2015 -0800
thread: use WaitForSingleObjectEx if available
Windows XP and up
100644 blob d58f74e5523dbc985fc531cf5f0833f1e9157cf0 src/utils/thread.c
100644 blob 840831185502d42a3246e4b7ff870121c8064791 src/utils/thread.h
vpx_thread: use CreateThread for windows phone
BUG=b/29583578
original webp change:
commit d2afe974f9d751de144ef09d31255aea13b442c0
Author: James Zern <jzern@google.com>
Date: Mon Nov 23 20:41:26 2015 -0800
thread: use CreateThread for windows phone
_beginthreadex is unavailable for winrt/uwp
Change-Id: Ie7412a568278ac67f0047f1764e2521193d74d4d
100644 blob 93f7622797f05f6acc1126e8296c481d276e4047 src/utils/thread.c
100644 blob 840831185502d42a3246e4b7ff870121c8064791 src/utils/thread.h
vp9_postproc.c missing extern.
BUG=webm:1256
deblock: missing const on extern const.
postproc - move filling of noise buffer to vpx_dsp.
Fix encoder crashes for odd size input
clean-up vp9_intrapred_test
remove tuple and overkill VP9IntraPredBase class.
postproc: noise style fixes.
gtest-all.cc: quiet an unused variable warning
under windows / mingw builds
vp9_intrapred_test: follow-up cleanup
address few comments from ce050afaf3
Change-Id: I3eece7efa9335f4210303993ef6c1857ad5c29c8
Use regular extended zero bin quantizer for both inter and intra
modes in the first pass. This doesn't affect lowres and midres
significantly, but would bring back 0.9% coding gains for hdres.
Change-Id: Ifa5977fa7b141fc5be595c0f3a4fc81a93f6606f
Originally the uniform quantization function was not being
replaced with the new_quant version in rdopt when new_quant
is turned on. This fixes the bug.
Change-Id: I593793bb909e1e1a6f89544eeca6783fe0576f25
1. Add "best_mv" in MACROBLOCK to store the best motion vector
during motion search, so that we don't need to pass its pointer
to various motion search functions.
2. Declare some functions as static when possible.
3. Fix some indents.
Change-Id: I0778146c0866cbc55e245988c59222577ea8260e
Use vpx_blend_a64_hmask and vpx_blend_a64_vmask to speed up
computing the obmc predictor. Clean up calc_target_weighted_pred.
Encoder speedup: 1.3%
Decoder speedup: 6.5%
Change-Id: I0c774fe53d22399e92a10d1daf3af0010d88d2c5
Fixed best error reported by loop filter selection, this value is used
during loop restoration to pick best mode. Baseline remains unchanged,
change in BDRate for loop restoration experiment:
-0.628 -> -0.625 for lowres,
-1.262 -> -1.283 for highres.
Change-Id: I69ef1608bc232b250ac46f59e31fdbed1a999dcd
- For experiment EXT_INTERP under high bit depth.
- Add unit test to verify bit-exact.
- Speed performance improvement:
On Xeon E5-2680, park_joy_1080p_12.y4m, 50 frames, encoding time
drops from 6682503 ms to 5390270 ms.
Change-Id: Iea4debf5414f3accf1eb5672abeab56a0539ac77
* changes:
vp10/encoder/rdopt.c: make a function static
vp10/encoder/rd.c: make a function static
vp10_convolve_ssse3.c: make some functions static
vp10/encoder/bitstream.[hc]: correct a prototype
vp10/common/idct.h: add some missing prototypes
highbd_quantize_intrin_sse2.c: add missing rtcd include
vp10: add some missing includes
Use vpx_blend_a64_hmask and vpx_blend_a64_vmask to speed up
computing the supertx predictor.
Decoder speedup of up to 4% has been observed.
Change-Id: I255a5ba4cc24f78dc905d25b6e2f7fbafac13253
- Made source buffers pointers to const.
- Renamed vpx_blend_mask6b to vpx_blend_a64_mask. This is more
indicative that the function does alpha blending. The 6, or 6b
suffix was misleading, as the max mask value (64) does not fit into
6 bits.
- Added VPX_BLEND_* macros to use when needing to blend scalars.
- Use VPX_BLEND_A256 in combine_interintra to be more explicit about
the operation being done.
- Added versions of vpx_blend_a64_* which take 1D horizontal/vertical
masks directly and apply them to all rows/columns
(vpx_blend_a64_hmask and vpx_blend_a64_vmask). The SSE4.1 optimzied
horizontal version now falls back on the 2D version. This can be
improved upon if it show up high enough in a profile.
- All vpx_blend_a64_* functions now support block sizes down to 1x1
(ie: a single pixel). This is for usage convenience. The SSE4.1
optimized versions fall back on the C implementation if
w <= 2 or h <= 2. This can again be improved if it becomes hot code.
Change-Id: I13ab3835146ffafe3e1d74d8e9cf64a5abe4144d
Directly call c functions, otherwise when EXT_TX is enabled, hybrid
transform other than combination of DCT/ADST has not been implemented, thus
will cause assertion failures in the switch loops in vp10_fhtnxn_msa() and
vp10_ihtnxn_nxn_add_msa().
BUG=webm:1239
Change-Id: I2379a07e5406f9489edcd2f3205682f679c9b091
vp10_optimize_b now takes between 40% to 60% of the TOTAL runtime
of the encoder, depending on bit-rate. It also contains 2/3 to 3/4
of the mispredicted branch instructions in the whole program.
Adding a few branch hints makes vp10_optimize_b around 2-5% faster
(dependig on bit-rate) when compiled with gcc/clang.
Change-Id: I1572733e18b4166bc10591b958c5018a9561fa2b
The combination of the two experiments improves the compression
performance gains:
lowres 2.5%
midres 2.1%
Change-Id: Id26c0a9474ce08893aa1d946365c7ff850fab57a
When the prediction residuals are all zero, reset the coeff rate
cost and the distortion value to be zero. This change doesn't affect
lowres set significantly, but improves several clips in the midres
set, like sintel_480p and mobisode2_480p, by a few percents. The
average performance for midres set is improved by 0.2%.
Change-Id: Idd5ebf2652e556a1b1c569fe3c48dacef3f11c32
Use int64_t type for distortion. This avoids integer overflow
issues in the trellis optimization function in high bit-depth
settings.
Change-Id: I550c3ca9f11a3191ef8638a152887018cd476141
test/assertion_helpers.h
test/randomise.{cc,h}
test/snapshot.h
Modfiy blend_mask6_test.cc not to rely on these.
Change-Id: I88b8933fe0a729a606797e5cd421795a544c612d
This reinstates the tests from commit
efda2831e5 with the appropriate
fixes for 32 bit x86 builds.
Change-Id: Ib331906c5b448ca964895ee9cbfd4266f67d1089
- Use int32_t instead of int in vpx_obmc{variance,sad} functions
- Remove weigthed_src and obmc mask strides and assume contiguous
buffers. These inputs can always be packed as contiguous arrays.
Change-Id: I74c09b3fb3337f13d39e13a9cb61e140536f345d
Originally we need to send the refresh flag and the virtual indices
mapping for the reference frame buffer update for show_existing_frame to
have the BWDREF_FRAME replace the LAST_FRAME.
To remove sending this information, we update the the virtual indices
of the reference frame buffer after the last_bipred_frame is encoded,
and therefore the decoder will receive the updated reference mapping
at the next non-show-existing frame.
As a result, we can save 4 bytes per show-existing frame, and get 0.12,
0.2, and 0.07 BDRATE improvement in lowres, derf, and midref test set
respectively.
Change-Id: I63d41ee6ea99884798f0778b789d2701e2f2d3e0
Reject ext-inter compound modes before doing full rate distortion
evaluation, if the corresponding single reference modes had a lower
modelled RD.
ext-inter speedup up to TBD.
Coding performance: TBD
Change-Id: I358bfb879c5ebe5e7afbf6f540cc784f8de14857
This offset value related to the bit depth has been taken care of
inside the function vp10_highbd_block_error.
Change-Id: I58dd8a53380ba4529d59837e56a951bc81a2962e
Commit 0d6980d7a1 removed some use
of the skip_txfm optimization, and the rest are not productive.
The current use of this optimization is only used with --good
and --cpu-used >= 3, however the overhead of this is higher than the
speedup it yields.
Removing this, and subsequently simplifying model_rd_for_sb yields
a net encoder speedup:
--cpu-used=0 ~1.5% faster
--cpu-used=3 ~2.0% faster
The code simplification is also significant.
Change-Id: I1dd668c32de15a2e912c59c42379d0f9e1032ff8
- Fix the over-writing bug in horizontal filtering as width = 2.
- Fix 10-tap vertical filtering which no longer reads one row of
pixel above the block.
- Fix 10-tap filter zero padding.
- Encoder speed slow down ~4.0%, compared to,
81ad953 Convolution vertical filter SSSE3 optimization
Change-Id: I9bb294a4529300081c29bf284e6bc6eb081cc536
Add the ability to pick between 3 quantization profiles.
The profile is chosen based on the entropy context at the
block level.
Change-Id: Iaea0485798441b7d635962c2563f3a477f582dac
This commit refactors the transform and quantization process for
sub8x8 blocks and unifies the related functions.
Change-Id: I005f61f3eb49eec44f947b906c4e308cab9935a2
Replace the expanded zero-bin quantizer with uniform quantizer in
the recursive transform block partitioning scheme. This improves
the compression performance by 0.4% for lowres.
Change-Id: I1c32ce9ebba0f0760e36a2c5bd20f2f5887ea5b4
- Apply 8-pixel vertical filtering direction parallelism.
- Add unit tests to verify bit exact.
- Encoder speed improves ~29% (enable EXT_INTERP) on Xeon E5-2680.
- Combinational cycle count of vp10_convolve() drops from 26.06%
to 6.73%.
Change-Id: Ic1ae48f8fb1909991577947a8c00d07832737e57
This fixes the unit test failure in the 1-pass settings of
EndToEndTestLarge.EndtoEndPSNRTest
bug=webm:1243
Change-Id: I7667c341f7c063f7ffb83786446bbbd1e498c1aa
Move the reference frame type definitions to common/enums.h file.
Replace hard coded numbers.
Combine repeated definitions.
Change-Id: I288e079a03e448014cc181bcdb3f88ee8ec8d139
Due to rounding used computation, HDB variance computation may produce
slightly negative values. This commit adds clamping to make sure
output variance values for 10 and 12 to be non-negative.
Change-Id: Id679aa55a4c201958c4c7d28cd8733b9246a71c8
This commit fixes the use of uninitialized context values in the
combination of supertx and var-tx.
Change-Id: I2d36badf5c9806ea402ce3e19515cc299e6b79e8
This commit refactors the reference frame structure used in the
dynamic motion vector referencing system, and makes it support
the bi-directional reference frames. This resolves unit test
failure (enc/dec mismatch) when both are turned on.
The compression performance (ref-mv + ext-refs) is improved by
0.2% for lowres.
Change-Id: I233624d8fccc1f69e82295f94de984ff056365dc
This commit adds an encoder workaround to support better
compatibility with a non-compliant hardware vp9 profile 2 decoder.
The known issue with this decoder is:
The decoder assumes a wrong value, 127 instead of the correct
value of 511 and 2047, for any assumed top-left corner pixel in
UV planes for 10 and 12 bit, respectively. Such assumed
top-left corner pixel is used for INTRA prediction when a real
decoded/reconstructed pixel is not avalable, e.g. when it is
located inside the row above the top row or inside the column
left to the leftest column of a video image.
Change-Id: Ic15a938a3107e1b85e96cb7903a5c4220986b99d
The tile boundaries should now be respected even between tile rows.
regardless of whether ext-tile is used or not.
Change-Id: I5a39fd274451114a4264215f97f12be2c908016d
When the next two states are identical, skip repeated cost table
fetch and multiplication operations. This makes the trellis unit
about 5% faster.
Change-Id: I0dbf7ad0a5732044e4e45dd59e9431a251c678f2
- Apply signal direction/4-pixel vertical/8-pixel vertical
parallelism.
- Add unit test to verify the bit exact result.
- Overall encoding time improves ~24% on Xeon E5-2680 CPU.
Change-Id: I104dcbfd43451476fee1f94cd16ca5f965878e59
decoding is done if the decoder is available, with errors handled
accordingly. the encoded frame count should be sufficient for this test.
+ remove HandleDecodeResult() as it's redundant given the base
implementation
BUG=webm:1233
Change-Id: I513c1c3475c58a746f4df627491bdc392fe21416
Properly restore the rate cost in the inner search loop of obmc
prediction. This avoids unexpected encoding behavior. It fixes
the unit test failure in obmc experiment:
AltRefForcedKeyTestLarge.Frame1IsKey/2
Change-Id: I667b219dfcf2f2c63d9d984900ed3cfd10c354bd
This commit refactors the trellis coefficient optimization process.
It saves multiplications used to generate the final dequantized
coefficients. It removes two memset operations on quantized
and dequantized coefficient sets. This improves the unit speed
by 10%.
Change-Id: I23f47c6e14582520a7f952f03ce8f72183e7f0e6
Each time a codec is enabled or disabled with the umbrella
--enable-vpN flag, set the encoder and decoder configurations as well.
This was done as a post-processing step but doing that lost the order of
the arguments.
BUG=webm:1205
Change-Id: Ic629bfdd06acc04bc5a7227309f36bba54dad8b1
This patch removed the experiment of BIDIR_PRED and merged the feature
into the experiment of EXT_REFS:
(1) Each frame now has up to 6 reference frames, namely
LAST_FRAME, LAST2_FRAME, LAST3_FRAME, GOLDEN_FRAME, (forward) and
BWDREF_FRAME, ALTREF_FRAME (backward);
LAST4_FRAME has been removed;
(2) First pass still keeps the 8 updates:
KF_UPDATE, LF_UPDATE, GF_UPDATE, ARF_UPDATE, OVERLAY_UPDATE, and
BRF_UPDATE, LAST_BIPRED_UPDATE, BI_PRED_UPDATE;
(3) show_existing_frame==1 is supported in the experiment of EXT_REFS;
(4) New encoding modes are added for both single-ref and compound cases,
through the use of the 2 extra forward references (LAST2 & LAST3)
and the 1 extra backward reference (BWDREF).
RD performance wise, using Overall PSNR: Avg/BDRate
Bipred only Prev EXT_REFS Current EXT_REFS with bipred
lowres: -3.474/-3.324 -1.748/-1.586 -4.613/-4.387
derflr: -2.097/-1.353 -1.439/-1.215 -3.120/-2.252
midres: -2.129/-1.901 -1.345/-1.185 -2.898/-2.636
If in vp10/encoder/firstpass.h, change BFG_INTERVAL from 2 to 3, i.e. to
use 2 bi-predictive frames than 1, a further improvement may be
obtained:
Current EXT_REFS with bipred
1 bi-predictive frame 2 bi-predictive frames
lowres: -4.613/-4.387 -4.675/-4.465
derflr: -3.120/-2.252 -3.333/-2.516
midres: -2.898/-2.636 -3.406/-3.095
Change-Id: Ib06fe9ea0a5cfd7418a1d79b978ee9d80bf191cb
Inter blocks that have SEG_LVL_SKIP active must be at least 8x8 in
size for bitstream conformance (see read_inter_block_mode_info in
decodemv.c).
This patch makes the variance based partitioning scheme stop at 8x8
blocks in inter frames. This satisfies the SEG_LVL_SKIP constraint
and is more in line with the original implementation of this function
(before it got extended for 128x128 superblocks).
BUG=webm:1234
Change-Id: I1fdd894569a9c0817713a77daabe4c8b8e1d00c0
Since combining VPX_DL_REALTIME with VPX_RC_FIRST_PASS is basically
nonsense, ignore the user's pass setting when this happens and
behave as if the requested encode is a single pass encode.
BUG=webm:1233
Change-Id: I5ee4c4e5838c4ca6d24988890aae490b10826db2
This commit takes the precise rate estimate for zero_token rate
cost update. It improves the compression performance:
lowres 0.15%
midres 0.23%
Change-Id: I36761079f75ce43c814f8c663667e359d4ac2cd4
The logic can be incorporated into configure.sh
Removes a dependency on ios-version.sh which was not part of DIST-SRCS
and removes a warning from 'make dist' sub builds:
../src/build/make/configure.sh: line 787:
../src/build/make/ios-version.sh: No such file or directory
Change-Id: Ic38314708eb278dd9d2a9769a670da32f6126637
The trellis optimization is going backward. Hence there is no need
to restore the token_cache values that is behind the current node
in the scan order.
Change-Id: I4da8a2e3f78bf9630e6667c85d8c387c5d94de9a
This value is signed in vp9/10
Cleans warning in Android build:
comparison of integers of different signs: 'unsigned int' and 'int'
if (cpi->frames_since_golden == (cpi->current_gf_interval >> 1))
~~~~~~~~~~~~~~~~~~~~~~~~ ^ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Change-Id: Ie137724982f3a46c8c1820548c1960d62a4e96f2
left_above_mv and above_block_mv return as_int
as_int is defined as uint32_t in vp8/common/mv.h
Cleans up -Wextra warnings:
signed and unsigned type in conditional expression
this_mv->as_int = col ? d[-1].bmi.mv.as_int : left_block_mv(mic, i);
^
this_mv->as_int = row ? d[-4].bmi.mv.as_int : above_block_mv(mic, i, mis);
^
left_mv.as_int = col ? d[-1].bmi.mv.as_int :
^
Change-Id: Ia043764e4ce93d2152d2269b1c7b28b5d5f814cf
This commit change to use int64_t to represent the sum of pixel
differences, which can be negative.
This fixes a number of ubsan warnings.
BUG=webm:1219
Change-Id: I885f245ae895ab92ca5f3b9848d37024b07aac98
Use ~15 instead of 0x..F0
Cleans warning in Android build:
comparison of integers of different signs: 'unsigned int' and 'int'
if (((cm->Width + 15) & 0xfffffff0) !=
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ^
comparison of integers of different signs: 'unsigned int' and 'int'
((cm->Height + 15) & 0xfffffff0) !=
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ^
Change-Id: Iac25839cde3425b7b9db7f33740dc46a551b7546
The test in arf_freq assumes any no-show frame as ALTREF_FRAME and
then calculate the minimum run between two consecutive ALTREF_FRAME's
based on this assumption. As BWDREF_FRAME is also a no-show frame and
the minimum run between two consecutive BWDREF_FRAME's may vary
between 1 and any arbitrary positive number as long as it does not
exceed the golden frame group interval, this test does not apply to
the experiment of BIDIR_PRED.
Change-Id: I70efb2c691fdc18601dbb8a7735ac2f27817e75a
Move the supertx skip bit and transform type past the recursive
prediction blocks. This is in preparation for using the segment level
skip feature for supertx blocks.
Change-Id: I8319414b0734144a9264e8a4a60940b6716b12a8
The old version used 64 bit loads, and then ignored the top half
of the result. This can cause asan failures if we read past the end
of a buffer. Switched to using 32 bit loads instead.
Change-Id: I57da127a26f869fb4b4f700b55408f6dc2fbbc1a
this file is shared between vp9 & vp10; this makes it available in the
presence of --disable-vp9
BUG=webm:1235
Change-Id: Iaf060c3c09afd2c7df69995b0c01589f78d4945e
For VBR: (1) allow newmv mode for golden ref to
select interpolation filter (as in last ref case), and
(2) don't use the more aggressive tx-skip testing logic for large blocks.
Only affects 1 pass real-time vbr mode (speed >= 5).
PSNR/SSIM metrics on ytlive set are all positive, ~0.5-2% gain.
Change-Id: I0ffbb0a9755563a5acd6230c58236e4f19a47266
This change is only for real-time mode if short_circuit_low_temp_var
is on. Add bias to last frame in choosing ref frame for partitioning,
when y_sad and y_sad_g are close. It speeds up real-time encoding by
0.5% on some clips with less than 0.1% overall PSNR drop on rtc test set.
Change-Id: I2a2110fe36455f3d8f0fc404aef2228f512e8df8
The current test case is only run for vp9 and vp10 when HBD
is enabled. This was mistakenly removed in:
d53f9a3 Enable VP10 HBD PSNR checking unit test
Change-Id: I88b8168ad1efd805d759238a037653a2901bf50d
This commit refactors the trellis coefficient optimization process.
It saves multiplications used to generate the final dequantized
coefficients. It also removes two memset operations on quantized
and dequantized coefficient sets.
The trellis coefficient optimization is on average running over
10% faster.
Change-Id: If3aa26d2a706c3012bf2b7ac059bf1825250e81f
This fixes some crashes in
VP10/EndToEndTestLarge.EndtoEndPSNRTest/ with high bit-depth and
ext-inter.
Change-Id: I10f0f08e1be4bd5c388616074d4aa3f91a2fda7a
This commit reworks the transform and quantization unit. It enables
the use of adaptive quantization for intra modes. This further
improves the compression performance:
lowres 0.36%
midres 0.79%
hdres 0.73%
The key frame coding performance is improved:
lowres 1.7%
midres 1.9%
hdres 3.3%
The overall coding gains are:
lowres 1.1%
midres 1.8%
hdres 2.3%
Change-Id: Iaec1a3a4c1d5eac883ab526ed076d957060479dd
Cleans warning in Android build:
comparison of integers of different signs: 'unsigned int' and 'int'
int n = (int)VPXMIN(sizeof(clear_buffer), data_end - data);
^ ~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~
Change-Id: I964355ceae6b39e22c0196294b25e28387f84945
Defined as unsigned in VP8_CONFIG
Cleans warning in Android build:
comparison of integers of different signs: 'unsigned int' and 'int'
if (cpi->oxcf.number_of_layers != prev_number_of_layers)
~~~~~~~~~~~~~~~~~~~~~~~~~~ ^ ~~~~~~~~~~~~~~~~~~~~~
Change-Id: I969e64cd2bfda6e61c564476dbd35b892b177646
The vpx_roi_map_t and vpx_active_map_t structures use unsigned rows
and cols but VP8_COMMON uses signed values for mb_rows and mb_cols.
Cleans warning in Android build:
comparison of integers of different signs: 'int' and 'unsigned int'
if (cpi->common.mb_rows != rows || cpi->common.mb_cols != cols)
~~~~~~~~~~~~~~~~~~~ ^ ~~~~
comparison of integers of different signs: 'int' and 'unsigned int'
if (cpi->common.mb_rows != rows || cpi->common.mb_cols != cols)
~~~~~~~~~~~~~~~~~~~ ^ ~~~~
comparison of integers of different signs: 'unsigned int' and 'int'
if (rows == cpi->common.mb_rows && cols == cpi->common.mb_cols)
~~~~ ^ ~~~~~~~~~~~~~~~~~~~
comparison of integers of different signs: 'unsigned int' and 'int'
if (rows == cpi->common.mb_rows && cols == cpi->common.mb_cols)
Change-Id: If1f118c20ffefd2530fbd371e6787cc8a6c31f0a
Mode is signed
Cleans warning in Android build:
comparison of integers of different signs: 'int' and 'unsigned int'
if (ctx->oxcf.Mode != new_qc)
~~~~~~~~~~~~~~ ^ ~~~~~~
Change-Id: I5cf81c40b103e688a31e1339511f5c9eb27edd38
Move filter intra modes search to the end, after regular
mode search.
On average no performance changes.
Change-Id: I9293c8fdf706ebf831fbd61c6bb81959790f4848
The speed feature sf->lpf_picl == LPF_PICK_MINIMAL_LPF is used
to disable loop filtering. This did not work with the loop-restoration
experiment, but now it is respected.
Note that this speed feature is only used in real-time cpu-used >= 8
settings.
Change-Id: I193723c9ac5f802ec31d8c8b4d37650796e065fd
mi->stride now depends on the maximum superblock size, and hence
the constant 8 padding is no longer appropriate. Traverse the array
using mi->stride instead.
Change-Id: I8e84b9fe1728f6663f8c10765fe32206375f1e71
Segment based loopfilter strength for supertx coded blocks is now
selected based on the minimum of all segment IDs within a supertx
coded block (same as the quantiser settings).
Change-Id: Ib056bd0d05f6a1d3b512a76deb4e2ad4db0f7dc4
Segment level quantizer settings for supertx coded blocks are now
selected based on the minimum of all segment IDs within a supertx
coded block.
This also fixes the 3 adaptive quantization modes with supertx.
Change-Id: Ib5db099539d4f82f240e1d745d6e5264f8b34cde
Exit early from function when supertx is used, rather than putting
the bulk of the function body in a single conditional.
Change-Id: I41f388a45bd46e4a6ee1c51f26782ed9bddff4e5
When using VARIANCE_AQ, we can change the segment assignment after
initialising the quantiser in set_offsets, so re-initialise it when
we do so.
Change-Id: I1f168553aaf0ade419f0d4bf05820cd591b87659
Explicitly signal when the segment map is being refreshed when
using VARIANE_AQ. This simplifies decisions about when the segment id
needs to be set from the previous segment map vs based on the current
variance.
Change-Id: Ieb12c950e9cfbc3f53f4d184880071dea805563c
This commit makes the dual filter experiment work with non-420
settings. It fixes unit test failure in EndToEndTestLarge.
Change-Id: I04f7afdee78f91389d9ff72947efa152098af930
Revisit the compression performance and complexity trade-off after
making the SIMD version of trellis optimizations. Before that,
reduce the transform-quantization function calls temporarily. This
would cause about 0.3% performance drop for lowres set.
Change-Id: I16917a6bd5c44ec6cd8cd0b59f3c336c4fd96dd2
1. Skip golden non-zeromv and newmv-last for bsize >= 16x16 if the
temporal variance obtained from choose_partitioning is very low.
2. Skip horz and vert INTRA mode for speed 8.
This change works best on the clips with little noise and with some
motion (e.g. gips_motion which has > 5% speed up). PSNR drop is 1.78%
on rtc test set, no obvious visual quality regression found.
Change-Id: Ib43b5b20e67809d03c5a6890818ddff59e1fc94a
This commit combines uniform quantizer with trellis based coefficient
level optimization. It improves the codebase compression performance:
lowres 0.8%
midres 1.0%
hdres 1.6%
Note that the current trellis optimization unit is using C code. This
will make the cost of the overall quantization process slower. A number
of optimizations will come up next.
Change-Id: Id441dd238e4844409d0f08f82604be777f3f5282
This experiment implements non-uniform quantization where
the width of the bins increases gradually to more closely
match a laplacian distribution of the coeficcients.
Performance Gain:
derflr: 0.15%
hevcmr: 0.675%
Change-Id: I25234244e3bcd94b87c1f77cf682190b61c8ef94
This reverts commit f19700fe52.
This crashes in SSE2/SumSquares2DTest.RandomValues/0 under x86 due to
alignment issues
Change-Id: I135d83ba6a7894c09d7c7a139b7eaf876416b40c
This reverts commit efda2831e5.
This commit causes segmentation fault at SSE2/SumSquares2DTest.RandomValues/0
Change-Id: I171937e4daf6f15323e8206418773deb03bd8c53
This test is failing when no experiments are turned on. PSNR is
31.96 when the threshold is 32.
broken since:
0d6980d Remove swap buffer speed feature
Change-Id: I3c29815b40d5282c37f52f4345b56992f8558b2e
Move initialization of a some new "twopass" values
to the function vp9_init_second_pass() and some other
small changes.
Remove #if GROUP_ADAPTIVE_MAXQ as this is always
enabled now.
Change-Id: I1dbec2fd7c419779848aa987c4cd7824d4df8456
the difference between src and dst will be signed, the error will be
unsigned.
quiets -fsanitize=integer:
unsigned integer overflow: 4294967295 * 4294967295
Change-Id: I580813093ee46284fde7954520dfcb1188f79268
the difference between src and dst will be signed, the error will be
unsigned.
quiets -fsanitize=integer:
unsigned integer overflow: 4294967295 * 4294967295
Change-Id: I502fd707823c4faaa7f587c9cc0312f057e04904
On scene-cut detected frames (i.e., high_source_sad = 1), use
nonrd_pick_partition (over choose_part + select_part), as
the nonrd_pick partitioning is generally better.
Small positive increase in metrics on ytlive set (~0.5 - 1%).
Negligle overall speed decrease, as its only used on scene-cut frames.
Only affects 1 pass vbr mode, speed = 5.
Change-Id: I07c89cbdc75f5bb16eb8e0e2773ead0980d2de5c
The assumption doesn't hold true in the current codebase. Remove
this speed feature to simplify the codebase.
Change-Id: I9b69f484c9b7cd612b825047cc5b2fce63ee0af7
The inter prediction residual can undergo different transform types
during the rate-distortion optimization search. The assumption used
in this speed feature no longer holds true. This commit removes the
related code to clean up the codebase and clear out unit test
failure in higher speed setting.
Change-Id: I7f7cd4df2345ed3e607c9fae75b38cd2dbde0cac
This reverts commit be12fefa4b
and commit 057c1c4034.
Also, the mismatch between the avx version and the
c version has been fixed.
BUG=https://bugs.chromium.org/p/webm/issues/detail?id=1168
For a rt encode using 1080p@60fps material, up to 11% performance
improvement overall was seen.
Change-Id: Icd1f216209ebc6fc0b8da885f32f356fa4355ed0
The eob of a block is not perperly set when skip_recode is true,
thus triggering assert(eob <= default_eob) to fail.
Change-Id: Ifecbe33dce2dc4903e0a80bd384dc09bf0dd8a44
If it's creating problems with some experiments, disable it under the
actual conditions where it doesn't work and file a bug.
Change-Id: Iab9f4bfe42ea926d49d371918da25f9a8938a20f
This commit re-works the transform type speed feature. It moves
the transform type selection outside of the coding mode loop. This
avoids repeated motion search if the best prediction mode is
chosen as NEWMV. It improves the speed performance for clips that
contain more motion activities.
For mobile_cif at 1000 kbps, this makes the baseline encoding 7%
faster and makes the encoding with dynamic motion vector referencing
scheme enabled 10% faster.
Change-Id: I93e2714b3e461303372c4b66a4134ee212faffd1
Code cleaup, use existing rolling_actual/target metrics instead,
set threshold to get same/similar effect.
Little/no change in metrics on ytlive set.
Change-Id: I74f3c3d0a143a9cf20dc9c3dee54c0f7e6a97a51
Add a max condition and lower the min value.
No change in behavior (metrics for yt live set) for the
default min/max_gf_interval=4/16 settings.
Small positive change when min/max_gf_interval=7/16
(for 60fps clips on ytlive set).
Change-Id: I1c1d72425c86c69419ea43fb9730130e81062f91
This patch will make sure the use of the BWDREF_FRAME for the
encoding of both the two types of bipredictive frames, namely
LAST_BIPRED_UPDATE and BIPRED_UPDATE. To realize it, the
updates on the cpi->ref_frame_flags have been moved to before
the encoding of one frame, instread of originally handled after
the encoding of one frame.
RD performance has been improved slightly, approximately by 0.17%
compared to before the applying of this patch:
lowres: Avg -3.474; BDRate -3.324
derflr: Avg -2.097; BDRate -1.353
Change-Id: I0aa19afd752293e345489fbff104c4351ca5498c
The segment counts are computed as part of packing the bitstream,
so they might have been computed already in the recode loop. Zero
the accumulator to avoid double counting.
This fixes some encoder/decoder mismatches.
Change-Id: Ib7816034cbbb1db41101116b706302b02fad3a2c
1. Wiener restoration filter now has normalization and evaluation of
quantization procedure.
2. Corrected scaling of bits in RD cost computation.
3. Changed dynamic range and number of bits for Wiener filter.
Observed gains: Overall 0.58% for low_res, 0.7% for mid_res sequences.
Change-Id: I8928b3ea493bfe1790926b00388d6c4bafc08e19
We can optimize wedge partition selection by pre-computing the
residuals of the 2 underlying predictors, and then blend these
to compute the sse of the compound predictor, without actually
having to compute and subtract the compound predictor.
Similarly we can pre-compute a proxy array which we can use to
cheaply check which mask sign would have lower sse.
Details are in wedge_utils.c.
Mathematically these are equivalence transformations, but due to the
finite precision the encoder output will be perturbed, though on
average this should make 0% difference.
ext-inter gains about ~4.5% speedup.
Change-Id: Ib2657c3209ae161b4090b58b4b6c392641bf2792
xd->plane[0].n4_h and xd->plane[0].n4_w are not set at that point
when using supertx.
While this fixes the immediate crash described in the referenced
bug report, there are still issues in the ref-mv experiment that
causes these tests to fail, so they are kept disabled.
BUG=https://bugs.chromium.org/p/webm/issues/detail?id=1230
Change-Id: Ibf8ef02847a903f8d10e6be28e16694db10c75af
add an upper bound to the framerate denominator above which 30fps will
be reported; fixes warning in corrupt / fuzzed files
Change-Id: I46a6a6f34ab756535cd009fe12273d83dcc1e9f1
The transform can only be skipped if both Y and U/V can be skipped, so
we always include the cost of tx size in the rate for Y. This will
get later subtracted if the transform is actually skipped.
Change-Id: I136a223e5596f18b69bb9f743e7e08438183a215
Provides more comprehensive coverage for --enable-coefficient-checking.
The intent is to make the --enable-coefficient-checking option
consistent with the VP9 spec.
Change-Id: I12d0120756d17572ca2b2d7e6a2ab9d8071d8d58
We used to cache the cost of the UV mode from the search with a
different previously tried Y mode, but the UV mode is contexted
on the Y mode, so caching the cost is inaccurate.
Change-Id: Ib003510afb6fc9befb7808b67b0be64f1c0a0804
Error messages:
..\vp9\common\vp9_loopfilter.c(1312): warning C4244: 'function' :
conversion from 'uint64_t' to 'unsigned int', possible loss of data
[.build-x86_64-win64-vs10\vpx.vcxproj]
..\vp9\common\vp9_loopfilter.c(1313): warning C4244: 'function' :
conversion from 'uint64_t' to 'unsigned int', possible loss of data
[.build-x86_64-win64-vs10\vpx.vcxproj]
..\vp9\common\vp9_loopfilter.c(1312): error C2220: warning treated as
error - no 'object' file generated
[.build-x86_64-win64-vs10\vpx.vcxproj]
Change-Id: Ia69260611997cd2ba41c7184a85ecead740a7c07
This patch factors in the different partition coding syntax used for
right and bottom edge blocks when doing RD search.
Change-Id: I2f31650512b6a4a7a2c03352414693aff6fbf87b
Increase in the damping used in adjusting the active Q range.
This does hurt rate accuracy a little in a few extreme cases
especially if the clip is very short*, but helps metrics.
* Note that the adjustment is applied at the GF/ARF group level based
on what happened in the last group. Hence for very short clips where
the length of a single group may be a significant % of the clip length
there is still scope for some drift that cannot be accommodated.
In practice most data points in our test sets are now much closer to target
than was previously the case with default settings and in some cases are
better even than they were with the command line undershoot and overshoot
parameter was set very low (e.g. 2%). For example in bridge_close at high rates
the old mechanism was unable to adapt enough to prevent extreme overshoot.
Change-Id: I634f8f0e015b5ee64a9f0ccaa2bcfdbc1d360489
Change to the calculation of the error divisor used in
get_twopass_worst_quality(). This follows on from other
changes to the rate control that impact the output of this
function.
Change-Id: I414fa9aa1e6a68a64dccea17c3712f44b8a0c10c
Changes to the function the redistributes bits from overshoot
or undershoot throughout the rest of the clip to respond more
quickly.
Change-Id: I90f10900cdd82cf2ce1d8da4b6f91eb5934310da
Added a factor based on the bit spend in the last arf group vs the
target to adjust the choice of the active worst quality in subsequent
groups.
Helps clips where previously there was a big overshoot or undershoot
to adapt and get closer to the target rate.
Change-Id: I67034b801679b99024409489a2273ea6fe23b8e6
The use of this value is preventing rate adjustment on clips
or sections that have very little motion but high noise and
this can give rise to some sections with massive overshoot.
Change-Id: I9a65c7c1148dc5d3a7d8b23e50fc1733f3661621
This is purely a refactoring patch and has no functional effect.
Uses of these masks can be arranged such that all input blocks are
contiguous in memory (stride == block width). In this case 1D versions
of operations can be used. 1D vector operations have superior performance
over 2D block equivalents as they are more processor cache friendly and
they can do away with a second loop overhead.
Change-Id: I2b76c9888aea2c857cc497e8a4b2841fd3dad54e
Replaced vpx_d45_predictor_4x4_ssse3(), vpx_d45_predictor_8x8_ssse3()
and vpx_d207_predictor_4x4_ssse3() with
created vpx_d45_predictor_4x4_sse2(), vpx_d45_predictor_8x8_sse2()
and vpx_d207_predictor_4x4_sse2() respectively.
It's mostly neutral or slightly worse than ssse3 in good cases and
better than ssse3 in the bad cases (but still worse than using the mmx
regs).
Change-Id: Ib0237ceb71d2c57b8a93fd3170330cfed9d56bdd
Use the same rounding method that is used throughout the codebase,
where the halfway value is rounded up rather than down.
Change-Id: I04e92850bc69a7d7a07b06e3d2ce97f6f2ada321
Skip intra-mode and some inter-modes (newmv, nearmv, nearestmv) for
golden frame if the variance got from choose_partitioning is very low.
Only for 1 pass real-time CBR mode and bsize >= 32x32, it has ~2.5%
speed up with less than 0.1% PSNR drop for rtc test set. Don't see
visual regression.
Change-Id: I70efbc95a1007231ae36f02c5b2fbf6cd35077ad
Reduce operations and jumps. perf shows CPU time reduced from 1.9% to
1.6% when decoding fdJc1_IBKJA.248.webm on Xeon E5.
Will apply the changes to vp10 after code review.
Change-Id: I9351509922855d8896ddef1ed093b3ca12619a61
Seperate prediction mode and tx type search for inter
modes. Enabled for speed >=1.
baseline:
speed increase 40%
compression drop 0.30%/0.29% on lowres/midres
ext-tx:
speed increase 160%
compression drop 1.08%/0.95% on lowres/midres
Change-Id: Ieb34b1ee80df6980d16e26a5783e08cc0deae55b
Add a speed feature to seperate prediction mode and tx type search
for intra modes: search for best intra prediction mode with fixed
default tx type first, then choose the best tx type for the
selected mode.
Coding performance drop:
baseline
lowres 0.10% midres 0.08% hdres 0.14%
with ext-tx
lowres 0.14% midres 0.25% hdres 0.20%
Speed improvement is 20% for baseline and 17% for ext-tx.
It is turned on for speed >= 1.
Change-Id: Ia5e8d39e8a4e2e42c521bfde938f8b6a98ab24f9
For non-rd pickmode:
best_pred_sad, computed for NEWMV-last, is only used for
skipping golden non-zero modes. Add condition to avoid this
computation if not used (i.e, if golden nonzero modes are not used).
And remove code for computing best_pred_sad for NEWMV-golden,
since that sad is not used.
No change in behavior; small speed gain (~1%) for svc encodes.
Change-Id: Ic2cbdef6c4e9a233a57c0db0eeac8ad5fcead366
This is for the bidir-pred experiment. Previously the length of the
bi-predictive frame group interval is fixed at 2, i.e. one
bi-predictive frame may be inserted every other frame. This patch
makes the length adjustable, i.e. any positive number may be
specified, but the use of the backward ref will be turned off if the
bi-predictive frame group interval is larger than the golden frame
group.
Further, an additional rate factor level has been added:
INTER_LOW
, which applies to LAST_BIPRED_UPDATE frames that are not used as
references.
Change-Id: I5514d34a64dd486bbb5756c2d0612946f598a789
convert the random value to int16 before subtracting 256 from it; quiets
a ubsan (sanitize=integer) warning
BUG=webm:1225
Change-Id: Ibc2c5a21f30e112bd6c180f7d6a033327c38d0df
Function level timing test shows about 27% time saving on
a Xeon E5-2680 v2 desktop.
Rename vp9_dct_sse2.c to vp9_dct_intrin_sse2.c for vp9 and
rename dct_sse2.c to dct_intrin_sse2.c for vp10 to avoid
duplicate basenames.
Actually vp9_fwht4x4_mmx/sse2() and vp10_fwht4x4_mmx/sse2()
are identical. TODO: They should be unified later if there is
no intention to keep a duplicate.
Change-Id: I3e537b7bbd9ba417c606cd7c68c4dbbfa583f77d
C does not allow for shifting into the sign bit of a signed
integer, and the two instances here become signed ints via
promotion. Explcitly cast them to unsigned MEM_VALUE_T to
avoid the problem.
BUG=https://bugs.chromium.org/p/chromium/issues/detail?id=614648
Change-Id: I51165361a8c6cbb5c378cf7e4e0f4b80b3ad9a6e
input_, ref_input_ and output_ were being allocated with new[] followed
by vpx_memalign, remove the former
Change-Id: Ia16d0f9b9317042a24445095ad3c284f4e7bb481
Followed the code style of other lpf fuctions.
These 2 functions put 2 rows of data in a single xmm register,
so they have similar but not identical filter operations,
and cannot share the same macros.
Change-Id: I3bab55a5d1a1232926ac8fd1f03251acc38302bc
- Avoid excessive copying
- Don't both searching if no update can possibly offer savings
- Simplify the interface
- Remove the confusing vp9_cost_upd256 macro
Change-Id: Id9d9676a361fd1203b27e930cd29c23b2813ce59
Major parts have been implemented as follows:
(1) Added BRF_UPDATE, LASTNRF_UPDATE, and NRF_UPDATE in firstpass.c;
(2) Added the handling for the scenario of
"cpi->common.show_existing_frame == 1" at the encoder;
(3) Added a new reference frame of BWDREF_FRAME;
(4) Have bwd-ref work with upsampled references.
Note that when the experiment of "ext_refs" turned on, this experiment
will be turned off automatically currently.
RD performance in Overall PSNR has been improved, compared against the
VP10 baseline:
lowres: Avg -3.312; BDRate -3.154
derflr: Avg -1.927; BDRate -1.176
midres: Avg -2.149; BDRate -2.001
hdres : Avg -0.567; BDRate -0.588
Change-Id: I4c06ff51cc20194bffbd4d2346e57ba3dcf6b62c
Apple's version format specification is strictly checked on app
store submission, even for embedded frameworks:
http://apple.co/1WgelY1
The build version number should be a string comprised of
three non-negative, period-separated integers with the
first integer being greater than zero. The string should
only contain numeric (0-9) and period (.) characters.
So that's room for "1.5.0" but not for "1.5.0-906-g656f9c4".
The full version returned from 'version.sh --bare' is now
embedded under a 'VPXFullVersion' custom key in the Info.plist,
so it can still be extracted from the resulting framework.
Change-Id: If34a58d02e407379d1f1859fda533ef7f983170b
vp9_diamond_search_sad_avx was disabled in:
057c1c4 disable vp9_diamond_search_sad_avx
this removes a missing prototype warning as the prototype is no longer
included in vp9_rtcd.h. the file can be restored if someone gets around
to fixing the issue.
BUG=https://bugs.chromium.org/p/webm/issues/detail?id=1168
Change-Id: Ia9fda4b81c53dc5fba7c31d780d761f886940b52
Removing redundant calls to memcpy from
build_wedge_inter_predictor_from_buf yields a net 4% encoder speedup
with ext-inter only. The output is identical.
Change-Id: If97d4e323a5c8aca90c84a25a72085e006b05446
This patch always compares the most recent show frames between
the encoder and the decoder to test the mismatch.
Change-Id: I68a91ad0996a598231450debfd616e24992419b5
This is to replace vp10/common/reconinter.c:build_masked_compound.
Functionality is equivalent, but the interface is slightly more
generic.
Total encoder speedup with ext-inter: ~7.5%
Change-Id: Iee18b83ae324ffc9c7f7dc16d4b2b06adb4d4305
Many codes require -mstackrealign flags. Although -mstackrealign has
been already added to CFLAGS of some modules, SIGSEGV occurs in other
modules than those modules.
The best way may be to find causes and to fix them. However, we
cannot know those causes until SIGSEGV occur really. In addition, if
SIGSEGV occurs in other programs, it will be fatal.
So adding -mstackrealign flags to CFLAGS unconditionally is
reasonable.
Change-Id: I999ef597a6afe97f5e7cc7bffaa866537c3eedd2
This reverts commit 2468163e07.
causes valgrind errors for overread of buffer in SubpelVarianceTest
Change-Id: I448e52c76f815ac199305b71f7d169f2bc167679
Move the logic for rechecking zeromv on denoised block out to simplify
the function. To simplify the param passing, add a new structure
VP9_PICKMODE_CTX_DEN which is only used when denoiser is enabled.
Change-Id: Iaa9b4396dfcb8147236c02d4a1868a09103a4476
This commit makes the filter extension in highbd aware of the
dual filter and ext-interp experiments to prevent enc/dec mismatch
when both experiments are turned on.
Change-Id: I11ac1f041bd5f73d61e839d6386d9c5d008da3f7
This commit makes the sub8x8 chroma component inter predictor
operate at 2x2 block level. This allows one to use the actual motion
vector associated with each individal pixel block. It improves the
compression performance
lowres 0.40%
midres 0.25%
hdres 0.15%
Change-Id: Ia40e07cc7fde463dbf660018850e024932136c4f
Use the same rounding method that is used throughout the codebase,
where the halfway value is rounded up rather than down.
Change-Id: Ie969ed7eb9fcc88a93a90c7e274fd82f336c7e4d
With ext-interp, a switchable interpolation filter is coded iff the
motion vector uses fractional pixel movement (ie, true subpixel
movement). With ext-interp and obmc enabled at the same time, the RD
search proceeds as:
1. Do motion search
2. Do interpolation filter search iff subpixel motion, otherwise use
EIGHTTAP_REGULAR
3. Evaluate obmc=0
4. Evaluete obmc=1 - This involves another motion search
If the motion search in step 4 yields an integer motion vector, while
the search in step 1 did not, then an interp_filter value other than
EIGHTTAP_REGULAR is invalid, and will cause an assertion failure
at output time, or a mismatch if not using --enable-debug.
The fix sets the interp_filter to EIGHTTAP_REGULAR if obmc=1 is picked
with an integer motion vector.
Change-Id: I4685d1ad537f41d833dc9eb64845956b67886cca
- Confirm input coeff buffer is 16-byte aligned.
- sizeof() prefer variable name instead of type.
- Fix function name (Capital first letter then Pascal case).
- Long base class name uses a newline (with colon and 4 space indent).
- Remove a unnecessary reference function variable.
- Method declaration precedes variable declaration in class definition.
Change-Id: I317f7e679926b5219f58c5f7d14512e94985e7fe
If a reference block is coded with sub8x8 block size, and if it
has sub-pixel level motion vectors, its prediction filter type
should be used as context information.
The coding performance gains of dual filter type coding scheme are
lowres 0.57%
hdres 0.88%
Change-Id: I68b98f2518d02f11c29d0256aeb45b2580fe5cac
This commit reworks the probability model contexts used in the
prediction filter type entropy coding.
Change-Id: I7abc68cb469248d0d7ca1046da3c086ecb7b066a
This commit clarifies integer value range for vairables used in
several variance functions, also change to use proper type
conversion to reflect the value ranges.
Change-Id: Ic3234b83a912ce1ad12d1b254f3378763e15cc5c
The inlining mirrors what was done with the low bit depth
inter_predictor. And the new highbd_inter_predictor name is more
consistent with other high bit depth functions.
Change-Id: I96437f745759aeec6260c6e39a974bf36f1c211c
- Integrate 5 flip transform types for each 4x4, 8x8, and 16x16
block, for experiment, EXT_TX.
- Encoder speed improves about 12%-15%.
- Update the unit tests for bit-exact result against C.
Change-Id: Idf27c87f1e516ca5b66c7b70142477a115404ccb
Use model for interintra mode search.
Speed-up about 5-10% with about 0.04 drop in efficiency.
lowres: -2.60%
Change-Id: I825bf0ba8a46eb7f19fc528c25b8df066fb8ea95
minus the non-existent nonrd portion. original change:
commit d642294b1c
Author: Jingning Han <jingning@google.com>
Date: Thu Feb 11 12:36:49 2016 -0800
Fix tsan error in VP9 sub8x8 intra mode search
This commit fixes issue 1141. The issue was triggered in multi-tile
encoding. The change properly saves and restores the block context
information in the real-time mode selection process. It removes
several redundant memcpy operations in sub8x8 intra block mode
search.
Change-Id: I35c9ad197f4bd500ec39b5fc833f052f19eee010
Change-Id: Idfa38c54c9e645479f6870d46f71fb1e91c071da
For the current stage, we assume a single prediction filter type
per direction in the settings of compound inter prediction modes.
Change-Id: I12a1afdd364b93fcee870bd11ad01fc40ab48cff
Increases number of wedges for smaller block and removes
wedge coding mode for blocks larger than 32x32.
Also adds various other enhancements for subsequent experimentation,
including adding provision for multiple smoothing functions
(though one is used currently), adds a speed feature that decides
the sign for interinter wedges using a fast mechanism, and refactors
wedge representations.
lowres: -2.651% BDRATE
Most of the gain is due to increase in codebook size for 8x8 - 16x16.
Change-Id: I50669f558c8d0d45e5a6f70aca4385a185b58b5b
The amount of border extension needed in the first stage inter
filtering is decided by the length of the second stage filter
kernel.
Change-Id: Icddbc58c02234d5df09ff0eeebcf166ffe689203
* changes:
Refactor and add flip unit test to vp10_inv_txfm2d_test.cc
Add flip feature to vp10_inv_txfm2d.c
add unit test for highbd flip transform
Refactor vp10_fwd_txfm2d_test.cc
Rename and change to how its updated.
Only affects 1 pass vbr.
Small change in metrics (< ~0.1%) on ytlive set.
Change-Id: Ibb1fe485699b6c4a8194951c8f229abe2f64b9a5
- Tx_type: DCT_DCT, DCT_ADST, ADST_DCT, ADST_ADST.
- Encoder overall instruction count drops 2.91%.
- Decoder overall instruction count drops 1.01%.
- Add unit test to test bit-exact result against C.
Change-Id: I908c9e0e5106c58f67dd72d28760e6c9ce54278e
Avoid accessing transform type that is not 2D-DCT if the transform
size is 64x64. This fixes an assert failure in this unit test.
Change-Id: I0dee865ea0925f5743b8a25c2f90eb6522b4d272
This commit unifies the inter predictor used in supertx at both
encoder and decoder sides. It removes the redundant decoder
implementations related to border extension.
Change-Id: I03985cee52604a518394232fa9258ce057af9c00
This commit unifies the inter predictor used in the encoder and
decoder side for super-tx experiment. This resolves an enc/dec
mismatch found in nextgenv2 nightly-run unit test.
Change-Id: I16ab8d6063edf9d2fba79473f470f1a592cc10a0
Also allows use of --enable-shared when configuring for Mac OS X,
producing a bare .dylib.
Enabling the shared framework bumps the iOS deployment target to 8.0,
the minimum required to support dynamic framework deployment in apps.
When not using --enable-shared, a static library for iOS 6.0+ will still
be built.
Minimum version settings have been moved into ios-version.sh so they
can be updated in a single place.
As with the static build, unless header search paths are manually
tweaked, users must add a VPX prefix on includes, such as:
#include <VPX/vpx/vpx_decoder.h>
A module map for headers is not yet included as inttypes.h is not
modular; this means that VPX cannot be used directly in Swift code,
but can still be pulled in through an Objective-C wrapper.
BUG=https://bugs.chromium.org/p/webm/issues/detail?id=1092
Change-Id: I28fb06ce65e48ed167a88c14a7bfb2861989317e
In the tile-coding experiment,
1. In tile decoder, added 2 set control APIs:
VP10_SET_DECODE_TILE_ROW and VP10_SET_DECODE_TILE_COL. It allowed
users to set the range of decoding at frame level.
2. Added a unit test while tile-coding experiment is on. It tested
both tile encoder and decoder to make sure the encoded frame
can be decoded as a whole frame or as independent tiles.
Change-Id: I73fd0632b685047cb9376008127cde72efa3fb2b
Fix the variable type mismatch between highbd_calc_masked_var_t and
the actual function definiton. This clears the related compiler
warnings in highbd with ext-inter experiment.
Change-Id: I0423318b16c867ed84700084ad21ca6e42edb321
In motion estimation stage for subpel motion, subpel variance is
computed use bilinear interpolation. The motion vector precision
used is at 1/8 pel and three bits are used to represent the x and y
subpel offsets. Based on this, the half pel check should be against
4, not 8.
Change-Id: I1f56fa1fa3f2f5e19a20d27983efe628557f170e
there are sse2 equivalents which is a reasonable modern baseline
Removed mmx variance functions:
vpx_get_mb_ss_mmx()
vpx_get8x8var_mmx()
vpx_get4x4var_mmx()
vpx_variance4x4_mmx()
vpx_variance8x8_mmx()
vpx_mse16x16_mmx()
vpx_variance16x16_mmx()
vpx_variance16x8_mmx()
vpx_variance8x16_mmx()
Change-Id: Iffaf85344c6676a3dd337c0645a2dd5deb2f86a1
Improves performance by 0.2%
lowres: -2.052% BDRATE
Also increases precision of the shift parameters (for further
investigation into different wedge shifts).
Change-Id: I59fcab9baa002e52a6487ed8d617185840a678ed
Added actual and absolute rate miss values to the opsnr.stt
stats output line.
Changes to the borg graphing may be needed before merge.
Change-Id: I1e9d548ce445d29002f0c59ebfd3957a6f15e702
Bug found by Yunqing relating to the correction for size at 8K and
above in get_twopass_worst_quality().
The basis for the correction was changed to the linear size relative to
1080P as a baseline and the adjustment has been clamped to prevent
problems at extreme images sizes.
For 1080P the results on our test sets were neutral but the low res and
mid res sets saw a small gain (0.1%-0.2% average).
I would also expect some gains on 4k and larger content where the
previous correction was overly aggressive.
Change-Id: I30b026b5f4535e9601e3178d738066459d19c8fb
Weighted single motion search is implemented for obmc predictor.
When NEWMV mode is used, to determine the MV for the current block,
we run weighted motion search to compare the weighted prediction
with (source - weighted prediction using neighbors' MVs), in which
the distortion is the actual prediction error of obmc prediction.
Coding gain: 0.404/0.425/0.366 for lowres/midres/hdres
Speed impact: +14% encoding time
(obmc w/o mv search 13%-> obmc w/ mv search 27%)
Change-Id: Id7ad3fc6ba295b23d9c53c8a16a4ac1677ad835c
Functions vp10_fwd_txfm2d_#x#_sse4_1 tested in this file
will be tested in vp10_fhts#x#_test.cc
Remove this to avoid duplication
Change-Id: Iaf21ab85b9a164fcf2a4574b3e13217e43b6255e
NEW_QUANT allows bin widths to be modified as a factor of the nominal
quantization step size. This adds functions to get dequantization
values based on the dequantization offset and 3 knots for a single
quantization profile.
Change-Id: I41f10599997e943cb3391c7a0847d8485b9d8b43
Add control API VP9E_SET_TARGET_LEVEL that allows the encoder to
control the output bitstream level and/or keep level related
statistics.
Usage:
255 do not care about level (default)
0 keep level related stats only
10 target for level 1
11 target for level 1.1
.
.
.
62 target for level 6.2
Usage for vpxenc:
--target-level=0/255/10/11...
Change-Id: I31d1aeca19358b893e7577b4e63748c8e614034a
Improves speed by about 10-15% by combining y-only rd with
modeling function in a better way.
Also, coding efficiency improves by about 0.1%
lowres: -1.805% BDRATE with ext-inter
Change-Id: I6ef1f8942ec6806252f3fcf749ae4f30dffe42b1
When not using ext-tile, there were still dependencies between tile
rows due to various tools (eg intra predictors) relying on the above
row or above mode info, which can be in the above tile. This is now
broken (the same way as it was when ext-tile is enabled) by fixing
the appropriate predicates.
Change-Id: I107dd0d8481775a792f14e05cfbbd761f16cdc1e
When constructing the intra predictor for rectangular interintra blocks,
the last row/column of the first square is copied back into the source
image (which is the current reconstructed image buffer) before
predicting the second square. The code used to use the height instead
of width for vertical rectangles, and vice versa for horizontal
rectangles, leading to overwriting the block on the right/below. This
leads to an encode/decode mismatch if the right/below block is in a
different tile and is encoded before the current block, which did happen
with multi-threaded encoding tests. This is now fixed.
Change-Id: I073a2a447a98b842b1394d72cc774a78cb296921
This change has no performance impact. It prepares the proper
function interface for better performance optimization.
Change-Id: I12e2f2deaf7f3adc603de0a74852116468c762f6
In VP10, REFRESH_FRAME_CONTEXT_OFF mode is only set when the error
resillient mode is on. Instead of being used to decide how to update
the frame contexts, it is used to decide if or not to reset the
frame contexts.
To verify, ran borg test on lowres set. The result is neutral.
Overall PSNR: -0.006%; SSIM: -0.006%.
Change-Id: Ic48265cf7488e80c6f5aab3eef7ba1c273506419
The original pruning function was not taking into account
that certain tx sizes/block sizes use a reduced tx set.
Prune 1: -0.3% performance drop, 20% speedup on foreman video
Prune 2: -0.48% perfomance drop, 30% speedup on foreman video
Change-Id: I557e919d97a89f787b47b3c8579a080db57f91d0
Without this patch, the experiment of ext-refs showed almost no coding
gains compared to the baseline. This is because when ext-refs is on, the
use of upsampled reference is off.
With this patch, the ext-refs experiment works with the upsampled
references and shows coding gains in Overall PSNR as follows, with ~5%
slow down for encoding time:
lowres: Avg - -0.965; BDRate - -0.844
derflr: Avg - -0.847; BDRate - -0.669
Note that the previous patch a912c6ec31
that "Make LAST_FRAME always point to the newly coded frame in ext-refs"
made ext-refs work with the upsampled refereces.
Change-Id: Id79248d71760109fb9198af4f45718b17455555f
For at least some of the implementations of sdx8f, such as
vpx_sad4x4x8_sse4_1, aligned moves are used to move the results into the
array.
Change-Id: I83df5a8e657b44e906d0d8b0bc154f1e5660f7f9
block_variance: This operates on 8x8s and would be safe with a int32 *
int32 to uint32 multiply, but this is potentially unsafe for 12-bit
input. Unfortunately the code already segfaults on 12-bit input:
https://bugs.chromium.org/p/webm/issues/detail?id=1223
calculate_variance: This operates on up to a 32x32 of 8x8s and can
overflow even with 8-bit input (log2((256*32*32)**2) == 36).
BUG=https://bugs.chromium.org/p/webm/issues/detail?id=1220
Change-Id: I1ca4ff6092db9a7580da371ee9a21f403fdadc40
- Tx_type: DCT_DCT, DCT_ADST, ADST_DCT, ADST_ADST.
- Update vp10_fht16x16_test.cc to do bit-exact test against
latest C version.
- HBD encoder speed improves ~1.8%.
Change-Id: Icfc799a212e5289bcf6cedcae3722032133a2bc6
Reduce factor for setting base-qp for active_best_quality (for inter-frames).
Small increase in metrics on yt live set.
Change-Id: I9cf0ac797783aeddbfaf1ff510696c9035d7c5ee
This commit fixes the compiler error in high bit-depth inter
predictor when dual filter type experiment is turned on.
Change-Id: I404a76a246477f2fcffc38a3275007d5dfe229cd
This change makes the c match the assembly and removes the todo's
associated with getting this to work.
Change-Id: Ie32e9ebb584a9d60399662d8bcb71b74fbd19d1e
This patch changes the encoder only for the ext-refs experiment. For
each newly coded frame to refresh the LAST_FRAME, the decoder is
notified that the LAST4_FRAME is to be refreshed, and read out the
updated reference frame buffer vitural indexes for the next coded
frame in a way that:
LAST4_FRAME => LAST_FRAME,
LAST_FRAME => LAST2_FRAME,
LAST2_FRAME => LAST3_FRAME, and
LAST3_FRAME => LAST4_FRAME.
Compared against the original ext-refs experiment in TOT, a small gain
is achieved in overall PSNR:
lowres Avg: -0.154
lowres BDRate: -0.044
Change-Id: I648810c146a3cd915b408274a9373b7d38324864
Make the bit-stream level support per direction filter type coding
for motion compensated reference.
Change-Id: I61a2360b301075f6734cfd9711b7ae68f214174d
These implementations rely on casting the pointers to load the data.
Clang implemented optimizations which automatically add alignment hints
to such loads. The 4x4 filters do not guarantee the necessary alignment
so the resulting assembly is broken.
https://llvm.org/bugs/show_bug.cgi?id=24421
BUG=webm:817
BUG=webm:892
Change-Id: I608885299f1f86ff83653b65e0e40d0ae87fb3fe
Use clear and correct type/function names.
Add ASM_REGISTER_STATE_CHECK wrapper for SSE4.1 function.
Conform macro EXPECT_EQ(expected, actual) convention.
Change-Id: I26c6430bea98a4fcb9727eb411b86a3b7abce933
* changes:
vp9_frame_scale_ssse3.c: make 2 functions static
vp9_pickmode.c: make function static
vp9_noise_estimate.c: make function static
vp9_aq_360.c: add missing include
vp9_idct_intrin_sse2: add missing vp9_rtcd.h include
vpx_dsp/*.[hc]: add missing vpx_dsp_rtcd.h include
The word 'pick' is usually used in functions that make decisions where
the bitstream allows multiple legal choices, and not to limit the
bitstream format itself.
Change-Id: Ia60709c29e004475e1aa8861aefded27ebaf4712
This experiment will implement the use of a backward prediction
reference without temporal filtering. No overlay frame will be
transmitted, instead, the flag of show_existing_frame will be turned
on.
Change-Id: I361a3004344e2ca6b63723f660635c0d790ee036
Makes the delta-qp stop little earlier on areas that have been refreshed enough.
This helps to reduce some pulsing artifact on noisy flat areas observed in some
noisy vc-clips.
Threshold changes only take effect for sources where noise level is estimated to
be >= medium level.
Only affects 1 pass CBR, non-screen content case.
Change-Id: Iacf557f6aa8abbcd6782c02ff2e6c14891960850
The encoder signals the interp filter type in the frame header if all
blocks use the same filter (see bitstream.c:fix_interp_filter). This
decision is made based on the counts, but with ext-interp, the counts
are actually only incremented for blocks that fail vp10_is_interp_needed
(see for example encodeframe.c:update_state), otherwise a default value
is used (EIGHTTAP_REGULAR). The decoder however first checks if the
interp filter is signaled at the frame level, and uses that filter type
for all blocks, even if the default value should have been used.
This patch makes the decoder first check with vp10_is_interp_needed
to see if the default value should be used and then checks the frame
level signaling, which reconciles the difference between encoder and
decoder.
Change-Id: I87857ade42dea06b0d5ec2a029e9219268334dbb
The test used to test that multi-threaded encode/decode resulted in
the same reconstructed image as single-threaded encode/decode. This
however did not mean that the multi-threaded encoder produced the same
bitstream as the single-threaded encoder, as the multi-threaded encoder
could use different forward probability updates and still produce a
bitstream that is sub optimal but yields the same reconstructed image.
The test now asserts that the bitstream is the same as well as the
reconstructed image. Also added more cpu-use values for testing VP10.
Change-Id: I324ed33a702c488b39e077f750d81a1ad1d7ea87
General code cleanup, but also use the same supertx condition for
ext-partition-types as for conventional partitions.
Change-Id: If86eb18b3c07b9c60434eec2c98b97ce93665b67
For 1 pass vbr mode:
Refactor to move the logic for gf setting based on up-coming
key frames to a separate function, so same logic can be used for
scene-cuts/changes.
Change-Id: Ic4ede308e08ba869bb62e4566e19ea31222c5229
Factor out common codes from vp10_get_pred_context_intra_interp().
This prevents a potential invalid access of pointers xd->left_mbmi
and xd->above_mbmi.
The coding statistics are identical.
Change-Id: I72dbf9380da7359b997bbe925010faab8e9e7f8d
Makes the noise estimation react little faster.
Little/no change in metrics.
Change only affects 1 pass cbr.
Change-Id: I13f0daa90ecbf9d49eb1cf2e48febd9d92292940
When building a dynamic framework with Swift compatibility, can't
include any headers that aren't in another module or you get an
error like this from Xcode on the including project:
Include of non-modular header inside framework
For some reason the system inttypes.h is not in a module, unlike
other standard C library headers... but it doesn't seem to be
actually needed on Darwin, so removing it doesn't appear to
be a problem.
Change-Id: I11d264483c54feefd9d2edf573afaef34ddcd0f2
When using git submodules, .git may be a file instead of a directory.
The -d test was failing in that case; switched to -e.
BUG=https://bugs.chromium.org/p/webm/issues/detail?id=1215
Change-Id: Iedf0e92bfeb003b28a415945dc729e6ce58c4fe4
"qc" in vp{9,10}_token_state is used to save quantized coefficients, this
commit changes the type from short to tran_low_t to properly reflect
the value range for highbitdepth build.
This fixes an out-of-range bug when optimize_b is used in highbitdepth
build.
Change-Id: Ibf330879e6ac6ae8f099e085caa9d3d9a889fde8
"qc" in vp10_token_state is used to save quantized coefficients, this
commit changes the type from short to tran_low_t to properly reflect
the value range for highbitdepth build.
This fixes an out-of-range bug when optimize_b is used in highbitdepth
build.
Change-Id: I914c6fd3d3f4b9d061f9ed7cc5f08a883ab59dcd
This is the set of 1D transforms that are used in each
ext_tx_used_inter set. The 1D sets will help speed up
the ext tx pruning functions.
Change-Id: Ib46ad26be2df60b3bfcd2f22d96e7f38ae286df5
This is an actual overflow where the result of the calculation is
materially changed, not just a negative value that is stored in an
unsigned.
Caught with fsanitize=integer on the VP9/AqSegmentTest.TestNoMisMatchAQ2/1 test.
Change-Id: I514b0ef4ae7ad50e3e08c0079aa204d59fa679aa
This ensures the multi-threaded and single-threaded encoder/decoder
always uses the same probability contexts.
Change-Id: I6f1e7c6bd8808c390c1dc0a628ae97db3acedf6d
In so doing this fixes a couple of bugs:
vpx_plane_add_noise.c needed to subtract a clamp instead of add.
And the assembly (mmx sse) had assumptions that parameters were
continuous in memory which was not true.
Change-Id: I76f2c43cf54bfc838eb2edf8a443eaaa7565d7b5
Decoding superframes correctly requires computing the end of the
frame contents in the bitstream precisely. This patch enables
ext-tile to do so.
Also extended superframe_test to test with multiple tiles if using
ext-tile.
Change-Id: I04bb8cde8755a3d764ee3c36aa8b7a6c5c9db742
Tile rows should now be independent, so make pbi->inv_tile_order
invert the decoding order of tile rows as well as tile columns.
This should improve test coverage. Also added more tile configurations
to the tile_independence_tests.
Change-Id: I14b0f2fa9241c1acaf9e2a07071952cb33feca77
With ext-tile enabled, the encoder test driver needs to configure the
tile sizes wit different values to encode using a single tile, and to
decode all tiles. This should fix most unit test failures.
Change-Id: I0a0d26737414669791f3bd8d80c537db09f06072
- iOS SDKs no longer ship with armv6 support.
- Our minimum iOS version means all target devices have neon.
- Remove armv6 darwin LD workaround.
- This removes a TODO.
Change-Id: I2fcb5b82c96213364275475be021c7dd8459d5c0
We enable this unit test to protect the high bit depth picture
quality from degrading under VP10 optimization and new experimental
tools.
Change-Id: I6297a44cf01954773e06549ab2a68c319fc848a8
Move skin superblock force split out of this function as well
as some minor code refactors. Checked bitexact for different speed
settings and different resolutions.
Change-Id: I6078cbe88dd9ce6c0b69470a8a0a8f8d2274161b
Fix 1: in ext-inter + obmc config, properly identify if the left
predictor used for obmc is a compound one in the case that the
neighbor uses wedgeinterinter pred and we will dump the ALTREF part.
This will fix the seg fault in unit test:
VP10/AltRefForcedKeyTestLarge.Frame1IsKey/0
Fix 2: in ext-tile + obmc experiment, handle the case that the
above block does not fit in the same row tile with the current one,
so as to prevent potential crashes.
Change-Id: I1c177d4f4ad15e10d11d8756e146496437753eea
This commit fixes an encoder segment fault in the codebase, when
the segmentation feature is turned on. The issue was introduced in
5cce322 Porting ext_partition experiment from nextgen
Change-Id: Ifb4c06c5a6976114a8bd061d40d0338a136abaaf
- Tx_type: DCT_DCT, DCT_ADST, ADST_DCT, ADST_ADST.
- Update bit-exact unit test against current C version.
- HBD encoder speed improves ~3.8%.
Change-Id: Ie13925ba11214eef2b5326814940638507bf68ec
this avoids the decoder test which was only correct for vp9, vp10 was
missed in the earlier change
Change-Id: Ib789c906d440c0e4169052cf64c74d5e4b196caa
Removes integer divides from backward updates for VP10.
Currently this is put in as part of the entropy experiment.
Coding efficiency change is in the noise level.
Change-Id: I5b3c0ab6169ee6d82d0ca1778e264fd4577cdd32
With ext interp, write_switchable_interp_filter calls
vp10_is_interp_needed, which needs access to the reference frame
buffers to check if they are scaled, the ref frame buffer pointer
at this point used to be uninitialized in the encoder resulting in
bitstream syntax mismatch when the encoder/decoder did not read/write
the interp filter element consistently.
Change-Id: Ie0be2a19cbfcb5639a751aa857458e91c23b8fe3
First, we only set use_4x4_partition for key frame where we don't
denoise; second, envision we have small partitions, we should pass the
actual block size to denoiser and make an early termination if needed.
Change-Id: I331f42046d792b17360723d17ff817d601394658
Wrap around behavior is enforced manually and we use the values in
arithmetic involving negative integers.
Change-Id: I199706b6f3af91f4fb6fe2ef302fbbc6d0cf5785
The product always fits in uint32_t, but the operands don't.
An optimizing compiler should generate the wraparound code.
(Verified with clang).
Change-Id: I25eb64df99152992bc898b8ccbb01d55c8d16e3c
ADL will look this up from the callsite namespace iff it is declared
before the callsite or from the parent namespace of the class type (the
global namespace).
This patch has been tested on MSVS 2015 and clang-3.8.
Change-Id: I00ba74712c9b617b9d81761abed1e14d8f25d8e3
Valgrind flags these up as needed by handle_inter_mode.
Initializing fixes some assertion failures in the unit tests with
only ref-mv enabled.
Change-Id: I4d56c356692745dbecd9f790cdbb8dbfbaf72d55
Remove the restriction that the neighboring predictor cannot be
used in obmc prediction if it is an interintra or wedgeinterinter
block. The inter predictor of the interintra block, or the first
inter predictor(using LAST or GOLDEN frame) of the wedgeinterinter
block will be exploited in obmc prediction.
Coding gain: 0.248% (2.833%->3.081%) lowres
Change-Id: I4ac0368b9d2f2956f266b30c1ac97db8bafa0742
The bug is introduced by commit 1a0352d, in which mv costs are
counted twice in joint_motion_search() in ext_inter experiment.
Change-Id: Ibace453df999d3c2e781d73f1f0912038fee2d4e
This commit enables 1/8 luma component motion vector precision
for all motion vector cases. It improves the compression performance
of lowres by 0.13% and hdres by 0.49%.
Change-Id: Iccfc85e8ee1c0154dfbd18f060344f1e3db5dc18
Reduce transform set for intra for 8x8 and smalller to 7 from 12.
Also fixes an issue with prob updates.
Enocder Speed-up about 8-10%
Coding efficiency very little change.
lowres: -2.996 (from -3.055 before)
midres: -2.482 (from -2.552 before)
Change-Id: I4ba50ff967521b33c748fe423bd92f7cf4105ebc
Block size passed into denoiser filter is always >= BLOCK_8X8 (in
vp9_pick_inter_mode), it is not necessary to check smaller block
size. Passed the bitexact test on clips with different resolutions and
noise levels.
Change-Id: I19fa3195d18c27d9e5de60dc11cff1522ef3714e
Fix will reset the consec_zero_mv map on non-skipped blocks with non-zero mv.
Adjust thresholds on consec_zero_mv in noise estimation and skin detection,
as more possible reset on map means lower thresholds should be used.
Change-Id: Ibe8520057472b3609585260b51b6f95a38fb777d
webm_read_frame is the only function now which requires
documentation for what the return value means (other two are quite
obvious - file_is_webm and webm_guess_framerate).
Change-Id: I7a4f7d8097b1d748812b2ee251ee718a0b5ce836
In VP9 internal denoiser, motion magnitude is computed from
best_sse_mv, which should be set to 0 at the begining. This bug may
cause visual aritifact in denoiser. Also, delete two improper comments.
Change-Id: I8710d2acba23320bc85cf72af17d65245c19438b
MMX and X87 floating point instructions cannot be mixed freely on
the 32 bit x86 architecture.
This fixes a lot of unit tests in the 32bit build with
--enable-ext-intra.
BUG=https://bugs.chromium.org/p/webm/issues/detail?id=1196
Change-Id: I0e1c3565f4b9cb4fc2d716e94d9c40e68b36fac8
Need to check that sse for non-zero mv has been set for the current block
(i.e., check that nonzero-mv is tested as a mode, so newmv_sse != UINT_MAX)
before forcing to not use zero-mv for denoising.
Also increase some thresholds (sse and sse_diff) for high noise case,
and use shift operaton instead of multiplication on a threshold computation.
Change-Id: Iae7339475d57240316b7fa8b887c4ee3c0d0dbec
- Optimization on tx_type: DCT_DCT, DCT_ADST, ADST_DCT, ADST_ADST.
- Overall encoder speed improves ~4.5%-6%.
- Update bit-exact unit test against current C version.
Change-Id: If751c030612245b1c2470200c9570cf40d655504
Resolved two TODO items.
Force a minimum value of 1.0 for frame duration as per section duration.
Column inactive zone is currently set to 0 as most of the serious issues
relating to inactive regions relate to letter boxing.
Change-Id: Ifbab3acf2c089d7305620a7ff7ed7c3536cc9235
In Aq mode 1 the segment and AQ delta for each block is based
on spatial variance. There may be a net imbalance between blocks
that have lower Q than the baseline value and those that have higher Q.
This patch monitors that imbalance and extends the allowed baseline
Q range for the frame to accommodate adjustment of that baseline value
to compensate.
Change-Id: Iae8a48c7c01fe2af94a141e149d03acf467237ca
This commit fixes an out-of-bound memory access case in the
loop filter mask setting. This issue was introduced in
10232ed Refactor loopfilter level arrays to 2D.
https://chromium-review.googlesource.com/#/c/336645/
Change-Id: I7101a4a79b9ecfdd8ec5ef13a0b314cc95f48d12
This commit fixes an encoding decision process issue that could
trigger enc/dec mismatch in the ext-inter experiment.
Change-Id: I6f10d1fd2fd1aa04e51df04c39a65cf72ac66c42
So it can be used even with aq-mode=3 not enabled.
Also cleans up some code in the places where its used.
No change in behavior.
Change-Id: Ib6b265308dbd483f691200da9a0be4da4b380dbc
Unit test shows manually developed SSE4.1 code would performs ~30%
better if TXFM_2D_CFG configuration is set in lower level. This
change only updates function signature. There is no performance
impact.
Change-Id: I62692bd50a21ffc8a944bbd6c155c0a2020ad77b
Removed this todo because of another todo which says none of this code
should exist. It should be integrated into the block by block encode
process as per the decoder.
Change-Id: I076bd15140a060e69c014dd7d7cd07fea260aba3
For 1 pass vbr mode.
Increase the gf interval for case where average Q is close to
max and high overshoot is detected.
Small increase in overall avg_psnr/sssim metrics (~0.2/0.1%) for ytlive,
but improves the low-end (low bitrate) for several clips (less overshoot).
Change-Id: Ifba40f25b4861b2e0d9832c82d5359a6a3dce9f2
This will facilitate bringing the zero node into the token set while
allowing its probability to vary independently.
Change-Id: I57b44c0fce44debb8e612021e44713b229d1b3cf
x->blk_skip used to be uninitialized (leftover from encoding the
previous block), if cm->tx_mode != TX_MODE_SELECT (which is used with
higher --cpu-used or --rt options). This resulted in degraded coding
performance when using cm->tx_mode != TX_MODE_SELECT.
This fixes the VP10/EndToEndTestLarge.EndtoEndPSNRTest/40 unit test.
Also fixed an edge effect where encode_block in encodemb.c used the
formal width of the block (without cropping at the right edge), to
look up blk_skip, while select_tx_block in rdopt.c used the cropped
width to set blk_skip.
Change-Id: I76d0f49ac5ab3ab54203573e0d7fcfcc1c6aa10d
x->blk_skip used to be uninitialzied (leftover from encoding the
previous block), if cm->tx_mode != TX_MODE_SELECT (which is used with
higher --cpu-used or --rt options). This resulted in degraded coding
performance when uning cm->tx_mode != TX_MODE_SELECT.
This fixes the VP10/EndToEndTestLarge.EndtoEndPSNRTest/40 unit test.
Change-Id: If39062927446798c626fc93694b4e6a4f35fa5da
This commit handles the zero motion vector residuals for single
and compound reference modes, respectively. It improves the coding
performance by 0.13% with no additional encoding complexity.
Change-Id: I16075a836025bd2746da2ff4698fb9261e4b08c1
We removed this adaption, which intended to reduce the size of
overlapped region if the neighboring block is a non-skip one. Thus,
now the width/height of the overlapping region is fixed as a half of
the current block.
Performance improvement (lowres/midres): 0.111%/0.102%
Change-Id: Ife75dad9d4eb355c78a05178b50cc015c442884f
More even spacing near key frame and avoid gf on scene cut
if its close to key frame.
Small increase in metrics for ytlive set (which uses key-period=150).
(~0.2% gain)
Change only affects 1 pass vbr mode.
Change-Id: If1e5a59baf1e0befbaf998522fbc47d94ac5b5df
Change only affects 1 pass vbr.
Use a q value somewhat larger (~6%) than avg_frame_qindex[INTER]
as basis for active_best_quality for inter-frames.
And use the minium of this (avg_frame_qindex) and the active_worst_quality.
This reduces some overshoot in ytlive clips.
Overall small but positive average increase in metrics (up on average ~0.2%).
Change-Id: Icdbaae7872d5675fd38a13c0ec6ce0e2e3b919ce
This commit re-arranges the transform type and size selectio
process. It removes an unnecessary rate-distortion cost computation
step. Local experiments show that this speeds up the encoding
process by 6% for both the baseline and the ext-intra experiment.
Change-Id: Iab3b86a63a1e9e55548466791ed5d29a0575c1e7
This CL fix the bug
rdopt.c:1687: choose_tx_size_from_rd: Assertion
`mbmi->tx_type == DCT_DCT' failed
It is caused by
1) mms register access before double operation
2) different compiler behaviors
code:
int64_t a = INT64_MAX;
double b = 1. * INT64_MAX;
printf("a < b: %d\n", a < b);
result:
a < b: 0
code:
--target=x86-linux-gcc
int64_t a = INT64_MAX;
double b = 1. * INT64_MAX;
printf("a < b: %d\n", a < b);
result:
a < b: 1
I remove the double operation and test it with EXT_TX experiment.
The psnr change is around 0.05%, which is considered as noise level.
Change-Id: If8935c70c8603617fcfa8571accd30ccdda786a0
In the rd loop, check the perf of obmc, whose mv is copied from regular
inter predictor, when wedge interinter is better than regular inter
(previously it will force allow_obmc = 0). The condition of the early
termination before this step is relaxed to avoid skipping too many obmc
predictions. The rates of the overhead are properly calculated for these tools.
The logic of the bitstream syntax:
(a single ref) the interintra flag is sent first, only if it is 0, we
send the obmc flag;
(compound refs) the obmc flag is sent first, only if it is 0, we send
the wedge interinter flag
Coding gain
lowres: 0.428% (2.287%->2.715%)
Change-Id: I5f3a34640b398e313cbf84235c9fe2073eb2173f
- Implemented Angie's new fwd txfm algorithm.
- Improve ~100% than last 64-bit version; 3 times faster than
original C code.
- Passed bit-exact unit test.
Change-Id: Ica30b9768706604a6d69fe42da778441f0f5f02e
With ext-ref enabled, it is possible that when trying to encode the
first true ALTREF frame after a keyframe, the previous ALTREF frame
(alias for the keyframe) is the same as one of the new LAST{2,3,4}
reference frames, and hence cpi->ref_frame_flags will have the ALTREF
bit clear, as computed by get_ref_frame_flags in encoder.c.
sf->alt_ref_search_fp forces the previous ALTREF frame to
be used as the only possible reference when encoding a new ALTREF
frame, but due to cpi->ref_frame_flags, some buffers will not be
initialized (see rdopt.c:7689 yv12_mb), leading to a segfault.
get_ref_frame_flags in encoder.c has been changed to prefer to keep
the LAST frame, then the ALTREF frame, then any of the LAST{2,3,4}
frames and then the GOLDEN frame in that order of preference in case
any of them are the same. This avoids the segfault and behaves the
same for the baseline.
Change-Id: I4da1991667614009da5d3061a6316c0d5dbc6c0c
This was never hooked up for the 32x32_34 case as the neon_asm version
in 3f7c12da, when the intrinsics version was added.
Change-Id: Ic7db4ce5850c637315f9fe9e2de93a4f8cf9e320
This avoids repeatedly checking the candidate motion vector
precision level at the decoder end. The compression performance
varies at 0.01% level.
Change-Id: I4a88e95decd900d0cac9a0c2e70ba43ef7ecac38
* changes:
Branch dct to new implementation for bd12
Change dct32x32's range
Fit dct's stage range into 32-bit when bitdepth is 12
Pass tx_type into get_tx_scale
Skip transform type search in modes with ref_mv_idx > 0. This
brings down the additional encoding time cost due to the DMR system
from 32% to 17%, at minimal coding performance regression.
Change-Id: Ie82e1d2831a313c6f1e47f7da221b51345023eb3
Fix several use cases where MAX_MV_REF_CANDIDATES is mixed up with
is_compound flag to avoid potential coding interruption.
Change-Id: Ifdee1ef8a81ef6d1c155315c6c6a3074aa7a8b5e
Change recursive weight for average_source_sad and
put some constraint on spacing between detected scene-cuts.
Change only affects 1 pass real-time mode.
Change-Id: I1917e748d845e244812d11aec2a9d755372ec182
Do not consider 4x4 transform when the maximum possible transform
size is 32x32.
Overall encoding speed is increased by more than 10%. Compression
performance is neutral on lowres, midres, and hdres.
Change-Id: Ifac61c3c9f4b0ab392bffd4d1faa373d91014cf1
The VP8_EFLAG_NO_UPD_LAST and VP8_EFLAG_NO_REF_LAST flags can be
passed to the encoder to signal that it should not update/reference
the LAST ref frame when encoding the current frame. With
--enable-ext-refs turned on, the new LAST2 LAST3 and LAST4 ref frames
could still be used or updated, which causes the
VP10/ErrorResilienceTestLarge.DropFramesWithoutRecovery/{0,1,2}
tests to fail.
With this patch, if --enable-ext-refs is used, then
VP8_EFLAG_NO_UPD_LAST and VP8_EFLAG_NO_REF_LAST also applies to the
new LAST2 LAST3 and LAST4 ref frames, as well as the LAST ref frame.
Change-Id: If482b1c09bbaf914eca8e0348a2367bff261661d
These caused the following warning with GCC 5:
warning: logical not is only applied to the left hand side of
comparison [-Wlogical-not-parentheses]
assert(!is_compound == (cm->reference_mode == SINGLE_REFERENCE));
Change-Id: If296aabb2311ceb7d903b395c1549ef81c2cbf9b
The minimal ans partition size is now one byte. This is checked in
ans_read_init().
The read_is_valid() condition is handled by setup_token_decoder().
Change-Id: I7b202b896630bc4285532208bf7cf84567afe158
- Interface function takes a local MxN function to call based on the
block size.
- Repetition call (w/o cache line miss) shows improvement:
~63% - ~340%.
- Overall encoder speed improvement: ~0.9%.
Change-Id: Ieff8f3d192415c61d6d58d8b99bb2a722004823f
Correct the setting of Q basis of GF/ARF in 1 pass vbr.
Existing logic would switch to using avg_QP of key frame if
avg_QP of inter is less than active worst (even if key frame is
not last frame).
Instead fix the logic (as per the comment) to use the lower of
active_worst_quality and avg_Q for inter as basis for GF/ARF
active_best_quality (unless last frame was key frame).
Increase in metrics: AvgPSNR/SSIM up by ~0.7/0.3 on ytlive set.
Change-Id: I9a628378ec6684bfda9457ebfc2384ef6d8579f7
Speed increase for ext-tx by 20% for a BDRATE drop of 0.26%.
The ext-tx expt becomes -2.66% BDRATE (reduced from -2.92%) for
the lowres set.
It turns out that reducing the set of transforms for intra from
12 to 5 makes very little difference in coding performance (~0.04%).
Most of the performance drop comes from the reduction is transform
set for inter. Currently there is a provision to control that with
a macro.
Change-Id: I7de05527bf72f96acc1e0ab8a74a849da0a141e5
Adjustment to stop excessive prediction decay triggered by blocks
or frames with extremely low spatial complexity which rendered the
comparison of intra and inter coded errors meaningless.
This was causing much shorter than expected groups on some 4k
test content.
Change-Id: I3f2c64200ef6dcef4721fc9f2ec09e480056ffc2
Uses a metric on fraction of smooth blocks derived from first pass
stats in a frame to adjust down the cq_level modestly in the cq mode.
The current implementation does not add much complexity, and is
fairly light in the adaptation.
Change-Id: Ic484e810d5bd51b7bb6b8945f378c7c3d9d27053
Update svm parameters with training data using new transforms
and remove DST from pruning functions.
Change-Id: I7bd1c4744455d571c1ecfb4cea14c25ac291f002
-Fix some bugs in row_scan and col_scan. In some cases, the above
or left neighbor was not considered even though it is available.
-When above or left neighbor is not available, try using the
top-left, top-right or bottom-left neighbor.
Compression improvement:
lowres 0.20%
midres 0.16%
hdres 0.20%
Change-Id: If521665589c7f29277b8e9223f21f4a8bf3fef39
Adjust the motion decay component to account for image size.
This has very little impact for smaller image sizes.
Average bdrate results for our HD test sets:-
Hdres set: opsnr +0,92%, Fast SSIM +1.6%
Netflix hd set: opsnr + 1.5%, Fast SSIM +3.1%
There are a couple of notable -ve clips such as cyclist and sunflower
which seem to be better with a shorter interval but also a few very big
wins such as Jets >12% psnr 22% Fast SSIM and from the Netflix
Netflix set PierSeaside 9.7% psnr and 18.2% Fast SSIM.
Change-Id: Ie43aaedaa74331ed83d624a13548094ac64fed9e
Change only affects 1 pass vbr mode, speed >=5.
Increase min_thresh, decrease boost, and set a min/max
value for gf_interval.
Change-Id: I9c1e1a1ab0c5780064eb62714ee39a72ea4d2107
Trap the case where we end up with a very short arf group just before
a key frame. Such a group often has poor quality and may cause pulsing.
For example if the KF is 17 frames away we are better doing two mid-size
groups of 9 and 8 than a group of 15 followed by a group of 2.
This becomes more and more important when coding with a short forced
kf interval though it may not impact our standard tests much.
Change-Id: I29d83d6637b203eac69be320dd35a7401a4678c1
The uncompressed frame header contains a bit to signal whether the
frame is encoded using 64x64 or 128x128 superblocks. This can vary
between any 2 frames.
vpxenc gained the --sb-size={64,128,dynamic} option, which allows the
configuration of the superblock size used (default is dynamic). 64/128
will force the encoder to always use the specified superblock size.
Dynamic would enable the encoder to choose the sb size for each
frame, but this is not implemented yet (dynamic does the same as 128
for now).
Constraints on tile sizes depend on the superblock size, the following
is a summary of the current bitstream syntax and semantics:
If both --enable-ext-tile is OFF and --enable-ext-partition is OFF:
The tile coding in this case is the same as VP9. In particular,
tiles have a minimum width of 256 pixels and a maximum width of
4096 pixels. The tile width must be multiples of 64 pixels
(except for the rightmost tile column). There can be a maximum
of 64 tile columns and 4 tile rows.
If --enable-ext-tile is OFF and --enable-ext-partition is ON:
Same constraints as above, except that tile width must be
multiples of 128 pixels (except for the rightmost tile column).
There is no change in the bitstream syntax used for coding the tile
configuration if --enable-ext-tile is OFF.
If --enable-ext-tile is ON and --enable-ext-partition is ON:
This is the new large scale tile coding configuration. The
minimum/maximum tile width and height are 64/4096 pixels. Tile
width and height must be multiples of 64 pixels. The uncompressed
header contains two 6 bit fields that hold the tile width/heigh
in units of 64 pixels. The maximum number of tile rows/columns
is only limited by the maximum frame size of 65536x65536 pixels
that can be coded in the bitstream. This yields a maximum of
1024x1024 tile rows and columns (of 64x64 tiles in a 65536x65536
frame).
If both --enable-ext-tile is ON and --enable-ext-partition is ON:
Same applies as above, except that in the bitstream the 2 fields
containing the tile width/height are in units of the superblock
size, and the superblock size itself is also coded in the bitstream.
If the uncompressed header signals the use of 64x64 superblocks,
then the tile width/height fields are 6 bits wide and are in units
of 64 pixels. If the uncompressed header signals the use of 128x128
superblocks, then the tile width/height fields are 5 bits wide and
are in units of 128 pixels.
The above is a summary of the bitstream. The user interface to vpxenc
(and the equivalent encoder API) behaves a follows:
If --enable-ext-tile is OFF:
No change in the user interface. --tile-columns and --tile-rows
specify the base 2 logarithm of the desired number of tile columns
and tile rows. The actual number of tile rows and tile columns,
and the particular tile width and tile height are computed by the
codec ensuring all of the above constraints are respected.
If --enable-ext-tile is ON, but --enable-ext-partition is OFF:
No change in the user interface. --tile-columns and --tile-rows
specify the WIDTH and HEIGHT of the tiles in unit of 64 pixels.
The valid values are in the range [1, 64] (which corresponds to
[64, 4096] pixels in increments of 64.
If both --enable-ext-tile is ON and --enable-ext-partition is ON:
If --sb-size=64 (default):
The user interface is the same as in the previous point.
--tile-columns and --tile-rows specify tile WIDTH and HEIGHT,
in units of 64 pixels, in the range [1, 64] (which corresponds
to [64, 4096] pixels in increments of 64).
If --sb-size=128 or --sb-size=dynamic:
--tile-columns and --tile-rows specify tile WIDTH and HEIGHT,
in units of 128 pixels in the range [1, 32] (which corresponds
to [128, 4096] pixels in increments of 128).
Change-Id: Idc9beee1ad12ff1634e83671985d14c680f9179a
This reverts commit 74aaa2389e. Unstable
under valgrind because of uninitialized reads. Limiting the bad bisect range.
Change-Id: I45b32f0ee0ba45795e7efb9947fb805830c8ce0e
In certain cases the code was subtracting the obmc cost
despite it not having been added previously.
For example with ref_mv, supertx, ext_inter, obmc & ext_refs
enabled the following test was failing but now passes:
"VP10/ArfFreqTestLarge.MinArfFreqTest/33"
Change-Id: I966853f34c18d5a1d4c7a56fa201c1b02973fc88
- Use arithmetic AND (&) instead of logical AND (&&) to
generate correct testing input.
- Fix variance reference function to be consistent with
our codebase implementation.
- Refer to the following issue:
https://bugs.chromium.org/p/webm/issues/detail?id=1166
Change-Id: I8c1ebb03e22dc9e1dcd96bdf935fc126cee71307
- Use arithmetic AND (&) instead of logical AND (&&) to
generate correct testing input.
- Fix variance reference function to be consistent with
our codebase implementation.
- Refer to the following issue:
https://bugs.chromium.org/p/webm/issues/detail?id=1166
Change-Id: I8c1ebb03e22dc9e1dcd96bdf935fc126cee71307
Avoid copy-block when denoising is at LowLow level (i.e., no denoising is done).
Instead, don't enter denoiser at all, and when level goes back up over kLowLow
do a reset in denoiser.
Change-Id: I0544adf58f4dd51ecc4a4607fcb0353bfbbb7a59
only output[0] needs to be set, store_output is more involved than a
movdqa in the high bitdepth case
Change-Id: I2cbd85d7cf74688bdf47eb767934fe42e02bff67
Avoid doing the mcomp in denoiser if we don't denoise the
block (because of motion/SSE/skin threshold, etc).
This can reduce encoding time (with denoiser enabled) by ~1.5-2%.
Change-Id: Ia699b68dfd37b89cdf3a82b8aa40e8c8f98a3d4f
Bitdepth 10/12:
Fit coefficient range into 32 bits
Fit codfficient * const range into 32 bits
Bitdepth 8:
Fit coefficient range into 16 bits
Fit codfficient * constant range into 32 bits
Change-Id: I50b5a3132e8a9f5155c971ab0f6eb52876d2b5ca
This make it more likely clean/low-noise content will
be set as LowLow, and hence no denoising will be done.
Also set early exit on denoising for small blocks.
Change-Id: I4a72bba3e6c5e2d523d304c39deacc9c39bf216c
Some cleanup and bugfix: pass mi_row/mi_col (not mv_col/mv_row)
to build_inter_predictors. This only affects case where
the frame is resized, but since denoising is not done on resized
frames, the fix has not effect currently.
Change-Id: I36617a7f0b43b6f49976745f15d400977e6ffa46
Switch to use new skin model.
And fix condition for denoising skin block.
Previous condition did not denoise skin blocks if the selected
mode was non-zero motion in current frame. Modify condition to
also force no denoising if that mode was not selected as zero motion
now and for at least "x" past frames in a row (x = 2).
Change-Id: I00753e3fe45b9a308a7ef43c58f11868e3bfc6b0
Use ANS for all entropy coded data in VP10 including the compressed header and
modes and motion vectors. ANS tokens continue to be used for DCT tokens.
Change-Id: Idf709a747150601e4d95d81ecfb3dc7253d349df
Decouples interintra modes and probability models from regular
intra modes, to enable creating/optimizing new interintra modes.
Also, fixes interpolation values for 128x128 interintra and obmc.
Change-Id: I5c2016db49b8f029164e5fe84c6274d4e02ff90e
Rename MI_BLOCK_SIZE.* -> MAX_MIB_SIZE.* (MIB is for MI Block).
Rename MI_MASK.* -> MAX_MIB_MASK.*
There are no functional changes.
This is in preparation for coding the superblock size at the frame
level, which will require some of these constants to become variables.
The new names better reflect future semantics, and hence make the code
clearer.
Change-Id: Iee08d97554cf4cc16a5dc166a3ffd1ab91529992
not strictly necessary, but allows projects using '-Wconversion
-Wno-sign-conversion' to reuse these headers.
Change-Id: Id1398d726c90173ccba9aea66798fcef6f20fa23
Fixes an issue with rectangular inter-intra blocks.
Includes various other refactoring and cleanups to enable fast mixing
of inter and intra predictors.
Uses only the best single inter reference so far for the inter-intra
search.
About 30% speed-up with a 0.1% hit in performance.
This is part one of overhauling on the ext-inter experiment. To be
continued in subsequent patches.
Change-Id: Id10ee100c78c6e00009a3a4f930a4435ef403a95
If --enable-ext-partition is used at build time, the superblock size
(sometimes also referred to as coding unit (CU) size) is extended to
128x128 pixels.
Change-Id: Ie09cec6b7e8d765b7555ff5d80974aab60803f3a
Amends previous commit to also handle subsampling correctly.
Change ID of prev commit: I6b07e6cf9b287ba4b5bd6599af4a7412e50b3bdc
Was causing occassional failures for 422 streams due to accessing
elements beyond the extent of the bmi array.
Change-Id: I37ebabf4c01ca84bcd1851428172bdf753805d98
Improve the readability in the related rate-distortion optimization
search control function of sub8x8 blocks.
Change-Id: I7f7456bf40a98aa5146abfe0488cda745b84d899
This commit makes the sub8x8 block to use its nearest neighbor's
motion vector as predicted motion vector for NEWMV mode. It improves
the coding performance by 0.12%.
Change-Id: I99e56715b327573ce7e8a26e3515a4984dadfd98
Moved the API patch from NextGen to NextGenv2 and also added this
API to VP10. An example was included. To try it, for example, run
the following command:
$ examples/vpx_cx_set_ref vp10 352 288 in.yuv out.ivf 4 30
Change-Id: Ib56bc3d365e530cfc8d859a13ddbf4c007907b81
This patch fixes 2 issues in Palette mode:
1. More memory is needed in PALETTE_BUFFER for 444 video format.
2. A merge issue caused by
https://chromium-review.googlesource.com/#/c/333940/7
Change-Id: I2aedc7dfdfb6b66fbd600189ec6e1e2cc6120d40
No need to do avoid shortcuts when all we are testing is the superframe
syntax. Decreases the run time up the VP10 version of the test from 22
seconds to 3 seconds on my machine.
Change-Id: If0c3551cbb8af8b803e02629e803e5f09da76cd1
- Wrote function: fidtx8_sse2() and fidtx16_sse2().
- Turned on vp10_fht8x8_sse2()/vp10_fht16x16_sse2() for new types.
- Updated 8x8/16x16 unit tests for accuracy/speed.
- Running 20K times with random numbers and getting through
tx type from V_DCT to H_FLIPADST, SSE2 speed improvement:
8x8: ~131%
16x16: ~66%
Change-Id: Ibbb707e932a08fec3b1f423a7dab280a1d696c9a
Change only affects 1 pass, vbr, speed = 5 (real-time mode).
Some improvement for high motion content.
AvgPSNR/SSIM metrics for ytlive set all up, on average ~2%,
some clips (high motion ones) up 4/5%.
Encoder speed down: on mynintendo_x1.1280_720.y4m: 47fps -> 44fps.
Change-Id: I9e3eaa6392dcb6b5b44ee6f43004f97ba859bc11
Skip checking obmc when regular inter predictor is not so good (the
rd-cost for Y residual is greater than the total rd of the best mode
so far.)
Performance change compared to full rd search:
+0.006% lowres, -0.056% midres
Encoding time :
1.14X baseline (was 1.42X)
Change-Id: I11350f955a20e1a2331be458537a915e09fbedf3
the vpx_decoder layer guarantees that when called directly this won't
receive NULL data and the reuse via decode() is protected by a NULL data
check and 0 size check (NULL data and non-zero data size is protected by
the vpx_decoder layer).
Change-Id: I7437fb5ca4e4aa431963d55b909d4d920f339be3
After porting tile coding from VP9 to VP10, some performance
degradation was seen because of the difference between VP9 and
Vp10 baseline. This patch disabled some features in VP10 while
tile coding is turned on. Also, an encoder control API was added
back for this use case.
Change-Id: I8f736db8388408a8cc35320a2f80abb02906571c
Skip filtered intra modes search in inter frame when DC mode is
worse than the best mode so far.
With ext-intra enabled, the overall speed is increased by 20~40%;
performance drop is 0.03% on lowres and 0.05% on midres.
Change-Id: I75d2503b067cf5e46e3533b97fb01497e125baa7
- Added function fidtx4_sse2().
- Turned on vp10_fht4x4_sse2() for these tx types.
- Updated 4x4 unit test for speed/accuracy.
- 4x4 Unit test passed.
- Running 20K times with random numbers for tx type from
V_DCT to H_FLIPADST, SSE2 against C, speed improves ~46%.
Change-Id: I828088b7f98dc0f5939a72e3fcd6cb0b8d8dd8bf
If configured with --enable-ext-tile, the codec uses an alternative
tile coding syntax in the bitstream. Changes include::
- The maximum number of tile rows and columns is extended to 1024
each.
- The minimum tile width/height is 64 pixels (1 superblock).
- A tile copy mode is added where a tile directly reuse the coded
data of a previous tile
- The meaning of the tile-columns and tile-rows codec parameters are
overloaded to mean tile-width and tile-height in units of 64
pixels.
- All tiles should now be independent, including rows within the
same columns, so large scale parallel, or independent decoding is
possible.
- vpxdec also gained the options to decode only a particular tile,
tile row, or tile column.
Changes without --enable-ext-tile:
- All tiles should now be independent, including rows within the
same columns, so large scale parallel, or independent decoding is
possible.
- vpxenc default tile configuration changed to use 1 tile column.
Change-Id: I0cd08ad550967ac18622dae5e98ad23d581cb33e
The mv is clamped in dec_find_mv_refs() to a smaller region
than the clamp in dec_find_best_ref_mvs(). See clamp_mv_ref
and clamp_mv2.
Change-Id: I47dd5f7fa8b42f2cc593559b4d7c782fe7bcb1db
- Use Makefile to control the build for highbd_fwd_txfm_sse4.c.
- Fixed hybrid transform (HT) types due to recent update.
- Added new unit test cases for highbd HT.
Change-Id: Ifd768a9b429a8c21ed40c1de8152fb5ac71e2f90
This commit separates the predicted motion vector from the nearestmv
motion vector in the coding process for both regular and sub8x8
block sizes.
Change-Id: I703490513b0194e6669ebf719352db015facb3e1
- Setup function vp10_highbd_fht4x4_sse4_1 for highbd SSE4.1
intrinsics optimization.
- Wrote SSE4.1 functions: load_buffer_4x4(), write_buffer_4x4(),
and fdct4x4_sse4_1().
- Used logic right shift to avoid coeff memory write/read.
- Turned on vp10_highbd_fht4x4_sse4_1 for DCT_DCT mode only.
- Improved overall encoding performance >2.3% for 50 frames
sequence, park_joy_1080p_12.y4m, in which, --input-bit-depth=12,
--bit-depth=12, 50 frames.
- Unit test passed.
Change-Id: Idd6dc6e472cbbf235f0ade4f66fbe859a860a004
In multi-thread case, the encoder may crash if using encoder option
tile-rows > 0. To prevent that, force tile-rows=0 in this situation.
This is a workaround for WebM issue 1095:
https://bugs.chromium.org/p/webm/issues/detail?id=1095
The further fix can be done by adding synchronizations after a tile
row is encoded. But this will hurt multi-threaded encoder performance.
So, it is recommended to use tile-rows=0 while encoding with threads
> 1.
Change-Id: I656cbcc200f8d0410d09530e7981ad8f32fe7bc9
This patch was to fix a reported Hangouts deadlock/freezing issue
in VP8 encoder(issue 27232610). The original encoder loopfilter
synchronization happened in the following frame, which was prone
to causing problems in some complex use cases. This patch simplified
the synchronization logic.
More testing needs to be done.
Change-Id: I38fd3f35d11f98fae1e44546aa5e4c6d6e19c4be
Allow the encode loop to select from a wider range of Q values
when encoding normal (non arf or kf) frames.
This change is targeted at improving psycho-visual quality in some
easy sections that are currently not getting enough bits.
This is likely to be a little worse from a metrics perspective and may also
have a small impact on encode speed in cases where extra recode
iterations are triggered.
Change-Id: I667eebf33c753bcbcf8b93596467369e5708b889
Adds a second threshold for recodes even on frames where
recode is normally disabled if there is a big rate miss.
Change-Id: Ifd4a34707da55ec15eb7cfb87de4644b8d76deb2
This has been ported under ext_partition_types because it is due
to be combined with the coding_unit_size experiment which is
already being ported under ext_partition
Change-Id: I47af869ae123ddf0aa99160dac644059d14266ee
Merge the functions that generate prediction by above/left predictors
for the encoder and the decoder.
Change-Id: I57e53a8f2eb8d3028c4ed0c9abdcbf00503f95a0
Decompose choose_tx_size_from_rd into three functions that determine
the transform coding rd at different stages. Besides the original
function, txfm_yrd() calculates the rd for fixed size and type.
choose_tx_size_fix_type() fixes the type and searches for the size.
It can enable other experiments to do restricted tx searches so as to
reduce the impact on speed.
Similar refactoring is done for select_tx_type_yrd() in VAR_TX.
Performance change in baseline is trivial:
0.014/0.001/-0.020 for lowres/midres/hdres.
Change-Id: I2ecbf6066329be088ec1bfb69013b657b14b8afe
This allows sharing more code paths with the rest of the code an allows
for easier compatibility with the other experiments.
Change-Id: Id288b533805a4d0657ec2f17542f2e6ad23ebdb4
Makes a set of 16 transforms total, adding all 1D
combinations of ADST and FlipADST, and removng all DST
transforms.
lowres, midres both improve by about 0.1% and hdres by
-0.378% in BDRATE but with fewer transforms that are also
simpler.
Further experiments to continue later.
Change-Id: I7348a4c0e12078fdea5ae3a2d36a89a319ffcc6e
Rework the interface to allow codec store the reference motion
vector list information for coding process.
Change-Id: I47e26587f6c0808655e4626f316ec7614a7ad8ed
This commit re-designs the probability model for the syntax elements
of the dynamic motion vector referencing system.
Change-Id: Icfb8203c7e8f64e10e99f5890e25e6f6b15fe5d1
This buffered ANS coder supports coding the symbols in forward (decode)
order. Rather than windowing or growing the buffer, right now this
coder merely asserts that the buffer will never overflow.
This approach should allow ANS to be used as a drop in replacement for
other entropy coders rather than requiring complicated reversal logic
throughout the codebase.
Change-Id: I6689271233d0e22fea94c51950415dad5af96598
This commit enables the dynamic motion vector predictor for NEWMV
mode. It allows the codec to select the best motion vector predictor
in a rate-distortion optimization framework for motion vector
residual coding. The compression performance is improved:
lowres 0.14%
midres 0.27%
hdres 0.24%
Change-Id: I6a601c74eb6cb0b71a613336d40363359f2edecd
Eearly termination if U plane RD cost is large enough.
No notable compression performance changes.
Change-Id: Ieeefc5859cb55d94391b502b4bd840bc8bcb2578
Fix the threshold for forcing refresh of golden frame based
on high motion. The current comparison was incorrect and
prevented this (force update of gf on high motion) from being used.
For now keep this logic under a flag (and off for now) so as to
not change behavior, until further testing.
Change-Id: Ib5f0082159a428b0603b9534e4bcb6f83e4ccb25
This is a cosmetic patch that removes a great deal of conditional
compilation around CONFIG_VAR_TX from the partition search function.
Change-Id: I9dcef9d4fe6847b793c77bdf565a5cacbdfacd59
This patch added two features to improve entropy coding efficiency
for coefficient tokens.
1. Choose 1 of 4 default probability tables based on q-index for
key-frames.
It is ported from nextgen branch:
https://chromium-review.googlesource.com/#/c/280586/
2. Do backward update after each superblock (64X64) row using
subframe token counts.
Coding gain: 0.1% on lowres; 0.42% on midres; 0.36% on hdres.
Much larger gain for key-frames: 2.6%, 2.3%, 1.7%.
Design doc: go/huisu-entropy
Change-Id: Ia3b6a615636be09247d70e4c520405637561532b
+5.857% BD-RATE on SCREEN_CONTENT
Leaving this off for non-screen content because:
+25.300% on TWITCH120
+37.833% BD-RATE on RTC
Change-Id: Ie0a312182d6cc859fb04298e4cd81d02b39e23fe
This commit fixes the computation of rate_nocoef for situation when
rate_y is uninitialized at INT_MAX for x->skip is true.
Change-Id: If3dde4e4ee16667f4408067d3bb3084f916272f1
In preparation for adding more 1D variants with ADST/FlipADST/etc.
BDRATE actually improves by 0.21% on lowres.
Change-Id: I2fa4720c69fe001fa666119a284dfc6b17fffab2
Optimized 2 up-sampled reference prediction functions in high-bit
depth case. This reduced the HBD encoding time by 3%.
Change-Id: I8663ffb5234f5e70168c0fc9ca676309fe8e98f2
Instead of testing all interpfilter-BMC/OBMC combinations, we choose
the best interpolation filter based on regular inter prediction.
Reduction in encoding time: ~10%
Drop in performance gain: 0.08% lowres, 0.04% midres
Change-Id: Ifc19097a918ac76b529db9af4c60e2c70e93f7ad
For 1 pass vbr mode: Increase the period of gf update on scene
cut (keep it same as orginal/default setting for now).
Change-Id: I679c3bd21152f6c4e486c8098d931c00e1d26b5f
Temporarily disable transform type selection for 32x32 transform
block size. This speeds up the encoding process. For bus at CIF
150 frames, the encoding time goes from 896s -> 762s (11% faster).
The compression performance for lowres set is improved by 0.15%,
and -0.029% for hdres.
Change-Id: If239b272970eb302150bec13b8cf192fbe045332
This is the identical change submitted for vp8 here:
https://chromium-review.googlesource.com/#/c/274107/
Tested this change on Mac OSX (10.10) and Linux
(Linux Mint 17 / Ubuntu 14.04) and in both cases:
- downloaded and compiled latest source for libvpx and ffmpeg
- confirmed ffmpeg would build sub-second frame rate webm files
via the previous patch
- confirmed ffmpeg would *not* build fps < 1 for vp9
- made this change, recompiled libvpn and ffmpeg
- confirmed ffmpeg would now create the same webm with
fps < 1
- confirmed the resulting file would play and was vp9 (e.g.
would not play in Firefox (Linux version complained it was
VP9 but mostly could play it) or older vlc, etc., but does
play just fine in Google Chrome and a newer version of vlc.
Sorry I didn't catch this last time - but this seems a solid
change and it's handy to be able to create frame rates
less than one second.
-jk
Change-Id: I38fa32148de8c4c359f228cf08b9a4b83b5a52fb
Using the up-sampled reference frames in sub-pixel motion search is
enabled as a speed feature for good-quality mode speed 0 and speed 1.
Change-Id: Ieb454bf8c646ddb99e87bd64c8e74dbd78d84a50
The change https://chromium-review.googlesource.com/#/c/329181/
also changed behavior for cbr mode, which causes some regression
in screenshare test in webrtc.
Resetting the specific change to leave the cbr behavior
unchanged for now.
Change-Id: I52df158806422f86398e1d2f522e92067d8325eb
Coding gain on screen_content is 12.2% (was 6.6%).
Some features such as frame-level color buffer, adaptive
entropy coding, are coming in future patches.
Change-Id: I2658cf5ec0cbb02cff685475759f3b68c9807697
Some adjustments to inter-mode selection for vbr mode.
Condition some of the bias to low/zero motion on cbr mode, and
don't use int_pro_motion_estimation for golden ref
(treat it same as last ref).
Change only affect 1 pass vbr mode, speed >=5 (non-rd pickmode).
Encoding time increase within ~5%.
Avg PSNR/SSIM on RTC set increase by ~2%, all clips up,
ranging from 0.5 to 4%.
Change-Id: I0048d0104a8816773d91a2b1484d601169d9bad7
This commit enables the hybrid 1-D/2-D transform coding scheme for
high bit-depth setting. It improves the compression performance of
ext-tx experiment by 0.98% for lowres_all set.
Change-Id: Ic27f5037f2c36b095a93b9f15dbae34bdcdf00aa
This commit enables the 1-D transform to use Manhattan grid vertical
and horizontal scan order for transform coefficient entropy coding.
Enabled in inter prediction mode, the hybrid 1D/2D transform coding
scheme outperforms the 2D-DCT based coding system used in VP9 by
lowres_all 1.7%
hdres_all 1.4%
As one coding option, in addition to the existing 17 other transform
types in ext-tx experiment, the 1D/2D hybrid transform improves
the coding gains:
lowres_all 2.2% -> 3.0%
Change-Id: I9cefa9d9e38224546d0afd67feecd9f8d4a16ab0
Don't advance the svc frame counters on dropped frame,
since this can break the referencing scheme and lead
to a crash/assert.
Updated svc-datarate unittest to add a lower bitrate test.
Change only affects 1 pass cbr svc, with frame dropper enabled.
Change-Id: Ibb7530b7a587a9344d46898d9286fd9e2ef0779c
- Implemented fdst16_sse2(), fdst16_8col() against C version: fdst16().
- Turned on 7 DST related hybrid txfm types in vp10_fht16x16_sse2().
- Replaced vp10_fht10x10_c() with vp10_fht16x16_sse2() in
fwd_txfm_16x16().
- Added vp10_fht16x16_sse2() unit test against C version:
vp10_fht16x16_c() (--gtest_filter=*VP10Trans16x16*).
- Unit test passed.
- Speed improvement: 2.4%, 3.2%, 3.2%, for city_cif.y4m, garden_sif.y4m,
and mobile_cif.y4m.
Change-Id: Ib30a67ce5d5964bef143d588d0f8fa438be8901f
Pixel domain distortion calculation is enabled for the rd loop of
inter sub8x8 and intra 4x4 cases.
Coding gain: 0.124% derflr, 0.122% derfhd
Change-Id: I43b47fe81b4f5ccc1c66bc626bd310c413a1ed87
- Inherited base class TransformTestBase to derived class VP10Trans8x8HT.
- Employed RunCoeffCheck() to test vp10_fht8x8_sse2() against C reference
function vp10_fht8x8_c().
- fdst8_sse2() related seven hybrid transform cases are covered in this
test.
- Test passed (4 test cases w/o EXT_TX; 16 test cases with EXT_TX).
Change-Id: Id9a9b308c707164a120d9ceb2c30e572026fb1d0
This commit enables a hybrid 1-D/2-D transform coding scheme and
the accompany entropy coding system. It currently uses hybrid
1-D/2-D DCT transform coding. It provides coding performance gains:
lowres_all 0.55%
hdres_all 0.43%
Change-Id: I2b30dcafd21eb2bb3371f6e854cbab440a4dfa78
For left side obmc, the input of the mask function is corrected as
the column coordinate.
Also, minor fixes for a compiler warning.
Change-Id: Ia981ef443d5b0285a93d73e5c7ab83f8c3a23464
Inherited class TransformTestBase to derived class VP10Trans4x4HT.
Employed RunCoeffCheck() to test vp10_fht4x4_sse2() against
C reference vp10_fht4x4_c().
fdst4_sse2() related seven hybrid transform cases are covered
in this test.
Wrote a header file for test base class. Some modification to
make sure the base class can be used for 8x8, 16x16, 32x32 cases.
All related tests passed.
Change-Id: I6b19a39d3ea30b657847781e78e73b829998a57a
This sets up the interface for 3 speed features that progressively
eliminate a greater number of transforms in ext tx using
pre-trained support vector machines.
Each speed feature still needs to be implemented.
Change-Id: Ia508aeadc0cffdc080fb227f357a5d1dfbca08e2
This commit fixes an encoding issue related to var-tx and ref-mv
experiments that causes the codec to use random values for transform
block skip flag.
Change-Id: I8daa6d6b88ea45b5bbeb81b43dd0eeff545c8e5a
Make the RANS implementation operate on cumulative distribution
functions rather than individual probability distribution functions.
CDFs have shown themselves more flexible to work with.
Reduces decoding memory usage from scaling O(num_distributions *
symbol_resolution) to O(num_distributions).
No bitstream change. This is an purely implementation change.
Change-Id: I4e18d3a0a3d37a36a61487c3d778f9d088b0b374
This allows the codec to use effective motion vector as the candidate
to produce the reference motion vector list.
Change-Id: Ib90be705fe28200c13376d6d7741800a61f13043
Use the superframe counter to set the key frame, and force
it to the key frame on base spatial layer only.
Also, update svc frame counters under frame dropping.
Update unittest: add specific tests with short key frame period.
https://bugs.chromium.org/p/webm/issues/detail?id=1150
Change-Id: I5b1c9a09253e6e5fbfce51b4cf603ae22d422b01
fdct16_sse2() was not bit-exact with C reference, fdct16().
The inconsistency was found by writing a unit test for
vp10_fht16x16_sse2(). Since the unit test needs a pending
change on the inherited base class. I will commit this unit
test after making a header file for this base class.
Passed the uncommitted unit test: vp10_fht16x16_test.cc.
Change-Id: If2b617883c633a3ea90c19e1d018240c8007102b
For 1 pass cbr mode: allow for two-stage 1:2 scaling
(which will use the 1:2 optimized scaler) if the spatial
layer is 1/4x1/4 of souce.
Without this change, the base layer for 3 spatial layers would
be using the non-normative scaler which is un-optimized/C code.
Change-Id: I9d73f92a4a96927d0f1d6bf75315c1e60513226a
The above-right and left-bottom pixels were sometimes not used even
though they are available. Results on lowres_all and hdres_all are
mostly neutral.
Change-Id: Ic13533dd498442ad5592b83bb5fabf053cc8e8f0
The sum of squared value of a block can overflow 32bit, this commit
changes to use int64_t to avoid the overflow issue.
Change-Id: I78fcd6999634f186f86d649cfce85d97a993d040
Use sharp filter to generate motion compensated reference for
temporal filtering. It improves the average coding performance of
VP9 speed 0:
derf 0.34%
hevcmr 0.38%
stdhd 0.58%
Change-Id: I1772a051be545de8c343055274e5ca0929d19cda
This commit back ports the fix from
https://chromium-review.googlesource.com/#/c/326940
It corrects the block partition context fetching in rate-distortion
optimization. It improves the average coding performance of speed 0:
derf 0.098%
hevcmr 0.102%
stdhd 0.282%
Change-Id: I8bcc6fe40ba5c6b50a6136daac116dcc738937ec
The double pointer in xd->mi handles this for us.
Cuts encode_suberblock()'s self time in half at rt speed 8.
Change-Id: I820dae24efdbf9a140bbeae82e4e2a5850317766
Up-sampled the reference frames to 8 times in each dimension using
the 8-tap interpolation filter. In sub-pixel motion search, use the
up-sampled reference frames to find the best matching blocks. This
largely improved the motion search precision, and thus, improved
the compression quality. There was no change in decoder side.
Borg test and speed test results:
1. On derflr set,
Overall PSNR gain: 1.306%, and SSIM gain: 1.512%.
Average speed loss on derf set was 6.0%.
2. On stdhd set,
Overall PSNR gain: 0.754%, and SSIM gain: 0.814%.
On hevchd set,
Overall PSNR gain: 0.465%, and SSIM gain: 0.527%.
Speed loss on HD clips was 3.5%.
Change-Id: I300ebaafff57e88914f3dedc8784cb21d316b04f
Fixes some issues introduced by a merge of two patches.
Also decouples the temporal interpolation filter from the switchable
filters for now for ease of experimentation with both separately.
Change-Id: If1c7c08adf00e0cf818fe8d0d3656c26ea65eb32
* changes:
x86/convolve.h: remove redundant check in FUN_CONV_2D
x86/convolve.h: replace while w/if for w < 16
x86/convolve.h: change filter[] || chains to |
restore the value for VP9 to 9999 to satisfy the current test
expectations; without this
VP9/DatarateTestVP9Large.ChangingDropFrameThresh/8 will overshoot.
Change-Id: I88dad574ae4ab10f923579824c7347ff468c7045
This reverts commit f51f0998e1.
This causes datarate tests to fail. Some are due to the new default
keyframe distance, another causes an assert even forcing 9999:
[ RUN ] VP9/DatarateOnePassCbrSvc.OnePassCbrSvc3SpatialLayers/0
test_libvpx:
vpx_dsp/x86/vpx_subpixel_8t_intrin_ssse3.c:853: scaledconvolve2d:
Assertion `y_step_q4 <= 32' failed.
Change-Id: I4ee4fea97f47e4f1a23b82a62e6afc6280961e38
Includes various cosmetic changes and refactoring including
naming the sharp filters differently (since they are no longer
8-tap).
Change-Id: Ida5a19ca0daa9f6a64a6734394c685b2a4a2564a
This commit unifies the encoder and decoder border extension and
motion compensated prediction process. Remove the decoder specific
flow to simplify the development flow.
Change-Id: I9c43bbe6d7c017e6da2db6a62c5bf3d0af7ccfce
The interintra experiment, which combines an inter prediction and an
inter prediction have been ported from the nextgen branch. The
experiment is merged into ext_inter, so there is no separate configure
option to enable it.
Change-Id: I0cc20cefd29e9b77ab7bbbb709abc11512320325
Reset the scale factors before build_inter_predictors.
Add datarate tests for 3 spatial layers, which exposed this issue.
Change-Id: I7f81efbe44345ecea9fdd5f639a4cca76aed3874
For 1 pass cbr mode: allow for two-stage 1:2 scaling
(which will use the 1:2 optimized scaler) if the spatial
layer is 1/4x1/4 of souce.
Without this change, the base layer for 3 spatial layers would
be using the non-normative scaler which is un-optimized/C code.
Change-Id: Ifcf526ec2aaf3e5fa7924588d9dd8660bf02fb46
some configurations may fail if AltRefTest is undefined though
VP8_INSTANTIATE_TEST_CASE is defined away.
Change-Id: I7272775a506718336bd6cee2225cf83bd72fede5
This commit uses 12-tap sharp filter to generate alter reference
frame. It improves the compression performance by
derf 0.45%
hevcmr 0.35%
stdhd 0.79%
No encoding time change is observed.
Change-Id: Ia5dc26d5aae6b9b0cb782e5a28dc5066eeeb2ec8
the same as vp8, with the same reasoning from:
2a0d7b1 Reduce the default kf_max_dist to 128.
see also:
https://trac.ffmpeg.org/ticket/4904https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=815673
+ restore vpxenc behavior of taking the library default rather than
forcing 5s
This change also exposes an issue with one-pass svc in cbr mode, keep
the old default in datarate_test.cc for now.
Change-Id: Id6d1244f42490b06fefc1a7b4e12a423a1f83e88
Implemented fdst8_sse2() function against C version: fdst8().
Added seven DST related hybrid transform types in vp10_fht8x8_sse2().
Replaced vp10_fht8x8_c() with vp10_fht8x8_sse2() in fwd_txfm_8x8().
Speedup: 18.1%, 11.5%, 22.0% based on speed test from
city_cif.y4m, garden_sif.y4m, mobile_cif.y4m.
Change-Id: Ia4aa1ea44c7a33e494f64ce843037f8703f975e3
* changes:
x86inc.asm: only set visibility for chromium builds
Only use .text sections for aout
Use .text instead of .rodata on macho
Copy PIC handling code from x86_abi_support
Set 'private_extern' visibility for macho targets
Expand PIC default to macho64 and respect CONFIG_PIC from libvpx
Use libvpx defines to set name mangling rules
Customize x86inc.asm for libvpx
Update x86inc.asm from x264
Adds hooks to use 32x32 ext-tx. Also adds scan orders for the masked
transforms for 32x32.
Make macro USE_MSKTX_FOR_32X32 1 in blockd.h to support 32x32 masked
transforms for ext-tx.
Change-Id: Ie6564830266651fcafae2d536c274dafd664ce17
These variable names were legacy from a previous version of this
function and in the current version they were confusingly backwards.
Change-Id: I4f6c1628f296fd5b650fd9c5e2d56d7daf66a3f6
This commit enables a context based motion vector entropy coding
conditioned on dynamic reference motion vector list. This (along with
the previous CL) imporves the coding gains due to dynamic motion
vector referencing based entropy coding:
derf 0.1%
hevcmr 0.2%
stdhd 0.7%
hevchr 0.4%
No encoding time change was observed.
Change-Id: I179c723844079195f6952a12582996a3ca9e9914
Instead of using model_rd_for_sb() to estimate the cost and make the
decision on bmc/obmc, we use super_block_yrd/uvrd() to calculate and
compare the real rd costs of bmc and obmc.
Average bit-rate reduction(%) of obmc experiment:
derflr/derfhd/hevcmr/hevchd
2.353/TBD/TBD/TBD
Before the optimization, the coding gain was:
1.582/1.109/1.600/1.164
Note: there is still some mysterious bug because that compared to
the previous version, the performance at low bit rate drops a lot.
Change-Id: I8dbee04a272190f10516a3953c1ae690f8136766
This commits adds a shift stage for FASTSSIM computaton when source
bit depth is different from working bit depth, to make sure metric
results are calculated in bit_depth consistent with source.
Change-Id: I997799634076ef7b00fd051710544681ed536185
This commit adds the ability to shift down the working buffer when
source bit_depth is different than working bit_depth. It does so
by shift down to be consistent with source bit_depth.
Change-Id: Idfdbfc614d73fe445d62e35e642cc7d75e9dc4ff
Don't initialize first pass costs for a number of symbols where first
pass probabilities aren't initialized.
As a side effect, an illegal read in the ANS experiment is fixed.
https://bugs.chromium.org/p/webm/issues/detail?id=1089
Change-Id: I97438c357bd88f52f5a15c697031cf0c3cc8f510
This commit unifies the motion vector cost buffers for full pixel
and sub-pixel motion search. The new motion vector coding system
provides 0.5% coding gains for 720p and above sequences and 0.2%
for lower resolution sets.
Change-Id: I927ec81eadc39d11a3c12b375221a1ddd2e8bf24
This commit extends the HBDMetricTests to handle testing for metric
computation where input source depth is different from working bit
depth.
Change-Id: I5d11101cc9603a3fd09e8439816bb982a0f1b654
Priviously, we do 12-tap interpolation even there is no sub pixel,
This could cause a bug becuase decoder doesn't extend border when there
is no sub pixel. In this situation, if we still do interpolation, we
will access the border extension which doesn't exist and cause a
memory error
Change-Id: I55b879722f0a10c5d13261bd9617a75c826a2418
This commit accounts for the context based probability model for
motion vector cost estimate in rate-distortion optimization.
Change-Id: Ia068a9395dcb4ecc348f128b17b8d24734660b83
This commit converts the scalar motion vector probability model
into vector format for later precise estimate.
Change-Id: I7008d047ecc1b9577aa8442b4db2df312be869dc
Use the existing scene/content change detection to better
update/adjust golden frame refresh.
Change only affects 1 pass real-time vbr mode, speed >=5.
Change-Id: I2963a5bb7ca4a19f8cf8511b0a925e502f60e014
This fixes a bug in HBD sum of squared error computation introduced
in #abd00505d1c658cc106bad51369197270a299f92.
Change-Id: I9d4e8627eb8ea491bac44794c40c7f1e6ba135dc
-Avoid unnecessary calculations
-Use SIMD when possible
Encoder is about 5% faster with the extra intra prediction angles
enabled.
Change-Id: I131056befe327cedab217ad4a40d5f2a11318acc
Preliminary tests indicated that these changes make cost_coeffs
approximately 20% faster which is a 2% improvement overall
Change-Id: Iaf013ba75884415cd824e98349f654ffb1c3ef33
this restores the previous version's behavior avoiding issues with
builds that may split sources on directory boundaries; protected
visibility may work in this case.
Change-Id: Ie759bd96c9ea5b45613f450dffa6e67eb45f5a8b
The read only sections are getting stripped on some OS X builds. As a
result, random data is used in place of the intended tables.
Change-Id: I4629c90d9e0ae4d4efc193a93be6fb93809ae895
Don't initialize first pass costs for a number of symbols where first
pass probabilities aren't initialized.
This brings a 1.22x first pass speedup.
https://bugs.chromium.org/p/webm/issues/detail?id=1089
Change-Id: I97438c357bd88f52f5a15c697031cf0c3cc8f510
This makes all metric computation to locate at some place, also gets
rid of duplicate code between vp9 and vp10.
Change-Id: I24a2707d183a2419cd18a8343010adae185ffcd4
Adds new 32x32 masked 1-d transforms that combine 1-D length-16
DCT with length-16 identity transforms.
To be continued in subsequent patches.
Change-Id: I0b4f66492d44c079b3c3b531ba48a97201de1484
This commit fixes an enc/dec mismatch in the dynamic motion vector
referencing experiment introduced in 837ef00.
Change-Id: I9fbe116fce118a80ef0f96bf41ce1f802547c2ee
replace with vpx_highbd_lpf_horizontal_edge_16 and
vpx_highbd_lpf_horizontal_edge_8 to avoid passing a count parameter
Change-Id: I551f8cec0fce57032cb2652584bb802e2248644d
replace with vpx_lpf_horizontal_edge_16 and vpx_lpf_horizontal_edge_8 to
avoid passing a count parameter
Change-Id: I848c95c02a3c6ebaa6c2bdf0983dce05cd645271
move to encoder_encode() as vp10_get_compressed_data() allocates data and
would require some modification to make its error return meaningful.
Change-Id: Ia5267c35d16ccd42b6da6d2136402b13e28f9159
move to encoder_encode() as vp9_get_compressed_data() allocates data and
would require some modification to make its error return meaningful.
Change-Id: I8ddc390a1441afd0ff937842fa4ad1053c956133
This bug made the rd loop use one-side obmc (compound of the current
predictor and the predictors of the left mi's, while the above ones
are ignored by mistake) to determine whether to use obmc. This fix
improved the compression performance by ~0.6% on different test sets.
Coding gain (%) of obmc experiment on derflr/derfhd/hevcmr/hevchd:
1.568/TBD/1.628/TBD
Change-Id: I43b239bedf9a8eebfd02315b1b036e140a998140
Add frame-level condition for reference masking: under external or
internal dynamic resize, allow for reference masking if none of
the references have been scaled.
Peviously, reference masking was turned off for the stream if dynamic
resize feature was enabled or an external resize event occurred.
reference_masking gives speed up with little/no loss in compression.
For speed 7 on rtc set: encoding time decreases by about 5-7%,
avgPSNR/SSIM goes down ~0.2%.
Change-Id: Ie4444577451ef954414d8fb4b2c99d65cadf1746
This commit fixes issue 1141. The issue was triggered in multi-tile
encoding. The change properly saves and restores the block context
information in the real-time mode selection process. It removes
several redundant memcpy operations in sub8x8 intra block mode search.
Change-Id: I35c9ad197f4bd500ec39b5fc833f052f19eee010
Removes the USE_DST2 flag that was on by default. DST2 performs
slightly better that DST1 and is faster to compute.
Change-Id: Ifb788f3f0a0e1995d7625230cec144b876f01206
Using this we can eliminate large numbers of calls to predict intra,
and is also faster than most of the variance functions it replaces.
This is an equivalence transform so coding performance is unaffected.
Encoder speedup is approx 7% when var_tx, super_tx and ext_tx are all
enabled.
Change-Id: I0d4c83afc4a97a1826f3abd864bd68e41bb504fb
External dynamic resize with swapping width and height was
not handled properly.
Fix is to re-init loop-filter under certain condtions.
Modify unittest to test this case.
Without this change test will fail.
Relates to: https://bugs.chromium.org/p/webm/issues/detail?id=1140
Change-Id: I7d81ca7fe0783b3bc103a52a7b7cf073a96be26e
In this experiment, an obmc inter prediction mode is enabled for
>= 8X8 inter blocks. When the obmc flag is on, the regular block-
based motion compensation will be refined by using predictors of
the above and left blocks.
Fixed some compatibility issues with vp9_highbitdepth, supertx,
ref_mv, and ext_interp.
Coding gain (%) on derflr/hevcmr/hevchd
OBMC:
1.047/1.022/0.708
OBMC + SUPERTX:
1.652/1.616/1.137
SUPERTX:
0.862/0.779/0.630
Change-Id: I5d8d3c4729c6d3ccb03ec7034563107893103b7f
Adds a wiener filter based restoration scheme in loop which can
be optionally selected instead of the bilateral filter.
The LMMSE filter generated per frame is a separable symmetric 7
tap filter. Three parameters for each of horizontal and vertical
filters are transmitted in the bitstream. The fourth parameter
is obtained assuming the sum is normalized to 1.
Also integerizes the bilateral filters, along with other
refactoring necessary in order to support the new switchable
restoration type framework.
derflr: -0.75% BDRATE
[A lot of videos still prefer bilateral, however since many frames
now use the simpler separable filter, the decoding speed is
much better].
Further experiments to follow, related to replacing the bilateral.
Change-Id: I6b1879983d50aab7ec5647340b6aef6b22299636
allocations done within this function are protected with
vpx_internal_error; adding the setjmp fixes a crash in
vp10_lookahead_push() under low memory conditions.
Change-Id: I5515017cd71b218840c506791b3a517da7ffc93e
allocations done within this function are protected with
vpx_internal_error; adding the setjmp fixes a crash in
vp9_lookahead_push() under low memory conditions.
Change-Id: I4b79dca37cc7fadc4b7633f0db44c0e406799bc6
This commit adds computation of PSNRHVS for highbitdepth build, it
also adds tests to make sure the calculation of psnrhvs metric for
10 and 12 bit correct.
Change-Id: Iac8a8073d2b3e3ba5d368829d770793212fa63b6
This commit aligns the rate-distortion metrics for both luma and
chroma components in super transform rate-distortion optimization.
It improves the coding gains due to var-tx and supertx experiments
by 0.2% for high resolution test sets.
Change-Id: Ib89d99e29cb5ee27b1f867e301954d4164d8b364
Brings the following commits to vp10:
269428e Tie the bit cost scale to a define.
d13385c Switch to 9-bit rate cost constants built on a 256 probability denominator.
ad43a73 Fix a signed overflow in vp9 motion cost.
1c9b091 Fix some interger overflow errors
fac947d Restore previous motion search bit-error scale.
Change-Id: I598ba7ee7efcde18439c31dfa96b86cbf297a580
An issue exists with reference_masking in non-rd pickmode for spatial
scaling. It was kept off for internal dynamic resizing and svc, this
change is to keep it off also for external dynamic resizing.
Update to external resize test, and update TODO to re-enable this
at frame level when references have same scale as source.
Change-Id: If880a643572127def703ee5b2d16fd41bdbf256c
This commit adds the computation of fastSSIM for highbitdepth build,
it also modifies the hbdmetric test to be more generic and applicable
for fastSSIM.
The 255 used for calculating ssim constants c1 and c2 is not exactly
scaled by 4x and 16x to 1023 and 4095, therefore requries the metric
test to have a thresold more tolerant than 0, currently at 0.03dB.
Change-Id: I631829da7773de400e77fc36004156e5e126c7e0
obmc: We add an obmc prediction mode at superblock level.
When it is enabled, predictors of the above and left blocks
are used to refine the regular block-based motion compensation.
Change-Id: I6310104ea3dfece16d736351e368861471dd1c9b
For dynamic resizing (whether the new codec size is determined internally
or externally set by user), we should for now keep rc.resize_allowed enabled.
This prevent the use of referene_masking for real-time mode
(in set_rt_speed_feature()).
Change-Id: Ibb7c3ff35be88afdf1a3c6db6693521766f177a3
Setting FIXED_TX_TYPE as 1 makes the encoder skip tx_type search,
about twice as fast.
This speed feature is off by defualt; we can turn it on when we
want to quickly test new ideas.
Change-Id: Ieab5807d17fcd54fce3e8ae2f59a18b42eb79408
to vp9_setup_pre_planes(), preventing the function
unscaled_value() from being called. unscaled_value()
returns the same value that was passed in. See
scaled_buffer_offset() in vp9_reconinter.h.
Change-Id: I2a6fbaf07972c2f212834929d29a2cbe72e399c3
This commit aligns the rate-distortion metric for the recursive
transform block partitioning and the super transform. It resolves
the conflicts between these two experiments. The coding performance
gains of the combined experiments (var-tx + super-tx) has been
improved:
derf 0.89% -> 1.9%
hevcmr 1.06% -> 1.8%
stdhd 0.29% -> 1.4%
hevchr 0.80% -> 2.3%
Change-Id: I7e33994ad70c1b2751435620815f867d82172f41
These costs are added in separately just before the computed
ref_costs_* are added in the calling functions, so they were
effectively double counted.
Change-Id: Ic941d0243460cc2e750791cfc508e97d8b90e8fd
The bit to error transformation got doubled as a result of going from
8-bit to 9-bit costs (change d13385c).
Use defines to derive the scale numbers and comment some of the fields.
derf: -0.023 BDRATE
hevcmr: +0.067 BDRATE
stdhd: +0.098 BDRATE
(These are substantially smaller than than the original gains from 8 to
9 bit costing.)
Change-Id: I6a2b3b029b2f1415e4f90a05709b2333ec0eea9b
When the codec frame size is the same as the reference frame size,
release the scaled reference before assigning it a new buf_idx.
Only affects 1 pass non-svc mode, where the scaled references are
release only under certain conditions (to prevent un-needed scaling
of the references every frame).
Modified a unittest that can trigger this bug without this change.
https://code.google.com/p/chromium/issues/detail?id=582598
Change-Id: I9a884e36ebd7608b1641ec2a469e20a4f829cf43
If the application changes frame size (external size changes),
and aq-mode=3 is on, reset the cyclic refresh.
Modify the TestExternalResize unittest (longer run with more resize
actions). Without this change an assert would be triggered on this
longer test.
Change-Id: I0eefd2cd7ffa0c557cca96ae30d607034a2599ce
This patch makes rd optimization use the same context for computing
the rate cost of coding the partitioning as the packer actually uses
when emitting it in write_modes_sb.
Change-Id: Idb1427bb2f9c37ab80c6aa182f7ff754ef0595cb
The value of use_highbitdepth flag is used for compute the size for
high bit depth buffer allocation, which should take value 0 or 1
depending on if the buffer is used for high bit depth or not.
Previously, the values is set to 8 or 0, this commit fixes the issue
and properly set the value for this flag to 1 or 0.
This cuts the size of highbitdepth buffer memory allocation to 2/9 of
the size prior to the fix.
Change-Id: I401518b5a6147e5d8a973e54f7ca6bc1892065e0
This commit enables entropy coding for dynamic reference motion
vector modes. The probability model is contexted on the ranking
categories of the reference motion vector candidates.
Change-Id: I09b58d98a409d63ec1a407331e29f8945b7ef17d
Fixes an issue where the tx_type was not set correctly for
sub8x8 inter and intra blocks. In the current syntax, for
sub8x8 blocks, there is still a single tx_type that is
transmitted. Ideally, this should be searched for the best
rd performance, albeit at the expense of encode speed.
For now, we just set it to DCT_DCT. Previously it was left
incorrectly as what was used for the previous non sub8x8
block.
derflr: BDRATE -0.277%
Change-Id: If76ba903bfbfd4d374cf1ac7d1daee50e92f0edd
Fixes an issue where the tx_type was not set correctly for
sub8x8 inter and intra blocks. In the current syntax, for
sub8x8 blocks, there is still a single tx_type that is
transmitted. Ideally, this should be searched for the best
rd performance, albeit at the expense of encode speed.
For now, we just set it to DCT_DCT. Previously it was left
incorrectly as what was used for the previous non sub8x8
block.
derflr: BDRATE -0.277%
Change-Id: If76ba903bfbfd4d374cf1ac7d1daee50e92f0edd
This commit enables the dynamic reference motion vector coding mode
for the compound inter blocks.
Change-Id: Ibe78fd8de6989db392cd67a9d81a69d680345ba1
Make this consistent with regular block size rate-distortion
optimization. It improves the compression performance:
derf 0.055%
hevcmr 0.129%
Change-Id: I112fe734f592c21bc7aa6efb7e3f269c4214ee7b
There were a number of compiler warnings:
1. int16_t to uint8_t in recon_intra.c;
2. double to float conversions in psnrhvs.c
3. intptr_t to int in quantize.c
4. size_t to int32_t in decoder.c
Change-Id: Id95423b17779dcfa6cf39d9a90fe8cb8b910f5df
Set USE_12_SHARP_FILTER to 1 to turn on the experiment
The psnr percentages increase
derf stdhd
lowbd +0.332 +0.318
highbd +0.476 +0.507
Change-Id: I783c0fc764ee8541645e100453c9b2073924e209
For 1 pass real-time mode. No change in behavior as only last
and golden are used as references in 1 pass real-time mode.
Change-Id: Ie4655014eee1a8b271542f29d74b2c6f7fed54c9
the results along the top and left border are then stored with a moving
window into the vector.
~40-67% faster on ARM, ~40-77+% on x86 depending on the block size.
Change-Id: Iab369aa2946a3ae4eb7290d512868fe5db92dbc8
delete apply_cyclic_refresh_bitrate(). unused since:
3472cbb vp9 aq-mode=3: Keep it on even at low bitrates.
Change-Id: I0fac9a31b59504e31000ac3a8f0b68e8d4320113
Temporaly disable warning for unused function for vp10, needs clean
out the warnings before re-enable the flag for vp10.
Change-Id: I5636f8cd607423f6ea6963db9c2cbd688e30b495
The definition is for the number of frames to check to determine the
recent decay rate, further to determine the next key frame in the
first pass of the encoder.
Change-Id: Ic696d6eb518a86fa296842273cf8767ef0b0e27a
when INLINE is defined and mips is not being targeted. otherwise keep
the old --enable-extra-warnings behavior
Change-Id: Iba576edbe5fca03efa56ce99eee11f9cafc573ad
Seperate the prediction angle search and fitler search.
It can reduce the computation overhead of filter search by as much
as 85%, while keeping more than 50% of the coding gain.
Change-Id: Id152f71e20ebcaca8b429bdd4ca1fbeb646fc6bf
-use larger threshold on y (as in vp8).
-add distance threshold for each cluster
-use larger skin distance threshold for first cluster
-add some early exist checks.
Keep default setting to model=0.
Change-Id: I1044b99ade4bb1f215a860a019a4d84cee2f7715
BD-rate performance improvement (on top of ext-intra):
derflr 0.22%
hevclr 0.36%
hevcmr 0.48%
hevchr 0.37%
stdhd 0.19%
Average speed impact on some derf clips is about 40% slower (on
top of ext-intra). Speed improvment is a to-do.
Change-Id: I8fe3fe8c5e4f60d0462778adbcc15c84dfbe7a25
This commit generalizes dynamic reference motion vector coding mode
to support multiple candidate modes in the rate-distortion
optimization scheme and to support the selection in the bit-stream
syntax. The maximum number of modes allowed is currently limited to
4. The syntax elements for the dynamic reference motion vector
modes are using binary codes. The scheme supports single reference
frame.
It improves the compression performance
derf 0.135%
hevcmr 0.098%
Change-Id: Id053d6ce76e8365e52727bd0d12d28ce3de2e0e8
It improves the compression performance of VP9 by 0.1% across all
test sets. No speed change is observed.
Change-Id: I59338c5c9e67bae22188f35fc3afbfe2a6bba6b0
It makes the encoder accounts for the block zero-forcing operation
when optimizing the mode decisions.
Change-Id: I2c8e243756080b446b8a53a9679f75c4c47148cf
Fixes assertion for football_422_4sif.y4m when supertx, var_tx and
ext_tx are all enabled. Problem was after subsampling, the u and v
blocks being encoded were no longer square.
Change-Id: Ie626f30a2e64538d33343a26d5124a79a6f2b985
This commit allows an adaptive motion vector referencing mode
approach. It checks the available reference motion vector candidate
list and decides the amount of motion vector referencing modes. The
current implementation assumes simple binary coding for the syntax.
The compression performance is improved by
derf 0.11%
hevcmr 0.38%
stdhd 0.09%
hevchr 0.23%
The coding gains due to the new reference motion vector system are
derf 1.0%
hevcmr 1.7%
stdhd 1.4%
hevchr 1.3%
Change-Id: Idf932fc373546fe59c8741f1b933ff656e8dbc3f
The postproc vp9_denoise() is a spatial denoise/blur function.
It was not intended to be used if temporal denoising is enabled.
Change-Id: I97d2dcb941e7cc49bbafce99d9286beb2693249d
Put check to avoid possible out of bounds when looping
over the blocks to estimate noise level.
No change in behavior.
Change-Id: I4b7b19b7edee0ae1c35b9dc0700b1bf9b304d7f5
* changes:
configure: extend armv7 hf target autodetect
configure: remove default CROSS for arm targets
configure: avoid default when CROSS is set to null
This commit changes SSSE3 optimized idct8x8 functions to work with
highbitdepth build.
With this commit and the previous one that enabled SSSE3 idct32x32
functions, tests showed virtually no difference on decoding speed for
file fdJc1_IBKJA.248.webm for the build with -enable-vp9-highbitdpeth
option and the build without the option.
Change-Id: Ibe0634149ec70e8b921e6b30171664b8690a9c45
Update blk_skip in update_state_supertx() and rd_supertx_sb().
Performance (SUPERTX+VAR_TX): TBD
(Eventually will merge update_state() and update_state_supertx())
Change-Id: I34ef982b80151ba2dfba745859cb2ca7b90dc888
This commit changes the SSSE3 assembly functions for idct32x32 to
support highbitdepth build.
On test clip fdJc1_IBKJA.248.webm, this cuts the speed difference
between hbd and lbd build from between 3-4% to 1-2%.
Change-Id: Ic3390e0113bc1ca5bba8ec80d1795ad31b484fca
the lookahead buffer allocation is deferred to receipt of the first
frame to allow profile changes. if the encoder was flushed before
supplying any frames the encoder would crash trying to dereference the
NULL buffer. vp8 is unaffected.
fixes mozilla bug:
https://bugzilla.mozilla.org/show_bug.cgi?id=1237848
Change-Id: Icee4b64de760476eee0d33b568f0a1010335ff13
Use multiple clusters instead of one and decrease
the distance thresholds.
Add a define to switch between models.
Default is set to existing (1 cluster) model.
Change-Id: I802cd9bb565437ae8983ef39453939f5d5073bb1
If a superblock contains alot of "skin" then force split
of 64x64 partition, and make some adjustments in mode selection.
This helps to reduce artifacts on moving face/skin areas at low bitrates.
Little/no change in metrics: avgPSNR/SSIM down by ~0.12%.
Small encoding time increase < 1%.
Change-Id: Ic57f52148c3716f391419fab0530d916e4c1d186
For aqmode=3, golden period update is set based on period of cyclic refresh.
Put a limit on max golden period update, for now set to 40.
And fix comment.
Change-Id: Icb61dd87c796cce2a5f5f7331c6a129540994696
Fixes integer pel MV usage for the sub8x8 case, which fixes a
rare mismatch issue.
Also adds some other minor missing code related to filter threshes.
Change-Id: I6b07e6cf9b287ba4b5bd6599af4a7412e50b3bdc
Reintroduce part of
Iaf2b717e6b8626b2b6a03226127221b776b49884
Which was later reverted in
I4c5b40ec63a6f19521191d3c730af87db3c4bc00
Change-Id: If3e5610ba3985ae7b4d952d8e616982465ac667a
There is still an assertion failure when tokenizing transform
coefficients for a supertx block, due to the eob not being set
consistently with the coefficients, so we always recode supertx blocks
for now. Also added further PICK_MODE_CONTEXT instances to avoid
potential clash between horizontal/vertical/split partition SUPERTX
trials.
Change-Id: I5f3da1fa0d8d20fc21face170487e1a285fd1cc6
The encoder did not update left_txfm_context and above_txfm_context in
MACROBLOCKD (used for choosing the probability context for the vartx
split bits) when the supertx bit was set for a block. The deoder on the
other hand did update these for supertx blocks. The encoder used these
to compute the context counts, which the packer then uses to adapt it's
probabilities. This results in the packer and the decoder using
different probabilities.
This patch harmonizes the encoder and the decoder by making the encoder
update the mentioned context for supertx coded blocks.
Change-Id: I3a22132124b1bce2ee501d640ceab374b19e3ca1
This is necessary when using SUPERTX, as the bitstream packer relies on
tx_size being set correctly to decide whether to output the block using
supertx or not.
Change-Id: I79e776b3b810f4a15b9dbc6afdd6fc90c73c8934
The loop filter relies on inter_tx_size in MB_MODE_INFO being set
properly when VAR_TX is enabled. Supertx coded blocks did not set this
previously at all, and the differing garbage values eventually resulted
in in a YUV mismatch between encoder and decoder after loop filtering.
This patch fixes this by setting inter_tx_size to the proper supertx
size in both the encoder and the decoder. This should also mean that
loop filtering is done at the proper transform boundaries, even when
supertx or vartx is being used.
Change-Id: I41a564cd6d34ce4a8313ad4efa89d905f5ead731
Limit oscilation detection in the case where overshoot is very very
large.
This keeps the 9-bit cost patch from breaking the DownUp reisze test.
The patch pushed us to an 11% undershoot right before a scene cut
causing a 1200% overshoot. (Whereas before we were undershooting by
only 6% before overshooting by 1200%).
Change-Id: Id90ccfab8aba872ccadc45b73b3bb097b895677f
Fixes some of the issues introduced by a merge from master.
derflr: -0.893% BDRATE
hevcmr: -1.667% BDRATE
Change-Id: I4c5b40ec63a6f19521191d3c730af87db3c4bc00
Combinations of different mv modes for two reference frames
are allowed in compound inter modes. 9 options are enabled,
including NEAREST_NEARESTMV, NEAREST_NEARMV, NEAR_NEARESTMV,
NEAREST_NEWMV, NEW_NEARESTMV, NEAR_NEWMV, NEW_NEARMV, ZERO_ZEROMV,
and NEW_NEWMV.
This experiment is mostly deported from the nextgen branch.
It is made compatible with other experiments
Coding gain of EXT_INTER(derflr/hevcmr/hevchd): 0.533%/0.728%/0.639%
Change-Id: Id47e97284e6481b186870afbad33204b7a33dbb0
In inter mode search skip all modes except NEARESTMV and DC_PRED.
10% less encode latency for large frames using the chromium remoting_perftests.
+0.313% BDRATE on the screencast set at speed -6.
Change-Id: Ib97a39dd8bcdeab545509e0e02d78ce7033f8c63
This patch fixes a couple of issues caused by change-id:
I15d20ce5292b70f0c2b4ba55c1f1318181481596
Changes to the code for when the ext_tx experiment is not enabled
were merged from master but as var_tx does not exist on master
the changes had not been applied to the case when var_tx experiment
is enabled
Change-Id: Iaf2b717e6b8626b2b6a03226127221b776b49884
Move it from vp10_adapt_intra_frame_probs() to
vp10_adapt_inter_frame_probs() because intra frames do not use
supertx.
Change-Id: I28c7391944848666054d4b990ac17a8ae08aaaee
Current implementation is a bilateral filter whose
parameters are transmitted in the bitstream.
derflr: -0.647% BDRATE
hevcmr: -0.794% BDRATE
This is a prelimary patch. Various other variations are to
be investigated next, that will hopefully be less expensive
on the decoder side.
Change-Id: I50634ae8f5014ad0bf7432306348908a349d81e1
Remove comment(s) and enable frame-dropper for tests.
Frame dropper for 1 pass svc was fixed a while ago:
https://chromium-review.googlesource.com/#/c/309230/
Change-Id: I5fd3192825b22e562db9210d3dc7b246a1799d8d
Make it consistent with the comment/intended behavior,
that is, only denoise if current block is zero_mv.
Change-Id: I3909761e802e80089752a493ab3646dc32698ded
Changes to mode selection for 1 pass SVC mode:
use base layer motion vector, changes to intra-prediction.
Change-Id: I3e883aa04db521cfa026a0b12c9478ea35a344c9
This patch changes the code for 16bit buffers to use the same
optimisation as is used for 8bit buffers. (See change-Id:
I0452da1786d59bc8bcbe0a35fdae9f623d1d44e1 for more information
about the optimisation)
Change-Id: I5f327a13a7b01fc356114a2aa9d1261bf76d8d69
This patch fixes a bug that causes the loop filter search to reset to
a low value or zero after each arf overlay frame. We expect the overlay
frames to need little or no loop filtering but this should not propagate.
Change-Id: I895b28474cf200f20d82793f3de40b60b19579fd
This is a pure-refactor in preparation to potentially raise the bit-cost
resolution.
Verified at good speed 0 and rt speed -6.
Change-Id: I5347e6e8c28a9ad9dd0aae1d76a3d0f3c2335bb9
More aggresive on avoiding denoising on skin.
May supplement this later by adding condtion onn consec_zeromv.
Change-Id: Ied92b332f9b24e821d2009f81d1565758588d9a5
NEW2MV is enabled, representing a new motion vector predicted from
NEARMV. It is mostly ported from nextgen, where it was named
NEW_INTER.
A few fixes are done for sub8x8 RDO to correct some misused
mv references in the original patch.
A 'bug-fix' for encoding complexity is done, reducing the additional
encoding time from 50% to 20%. In sub8x8 case, the old patch
did motion search for every interpolation filter (vp9 only
searches once). This fix also slightly improves the coding gain.
This experiment has been made compatible with REF_MV and EXT_REFS.
Coding gain (derflr/hevcmr/hevchd): 0.267%/0.542%/0.257%
Change-Id: I9a94c5f292e7454492a877f65072e8aedba087d4
ext-partition: to hold partition extensions (ex. ext-partition,
ext-coding-unit-size from nextgen)
loop-restore: to hold in-loop restoration filter (ex. loop-postfilter
from nextgen and other Wiener restoration filters)
Change-Id: I71c7f1588f05fb0f2b00f7004a78e90c9cceae3f
Fixes a breakage introduced with the latest merge from master and
cleans up a couple of compiler warnings.
Change-Id: Ia55b39ba78e43f6fe52c54d7f34faa4dd6bbbf26
This commit considers the case where a single reference motion
vector pair is found in the candidate list. It treats this pair
as the effective motion vector for nearestmv mode. This improves
the coding performance by 0.06% for stdhd sets.
Change-Id: I9ce12f456b52912933e05c18c3841a78c26155d2
Different quality levels are used for different regions in
the frame depending on how far they are vertically from the
center. Specifically, three segments are used based on the
mi_row index with respect number to the number of mi_rows in
the frame.
Change-Id: Ifc8b777bc58ea8521dffc4640360c67d99f8d381
This reverts commit ea48370a50, reversing
changes made to 15939cb2d7.
The commit was insufficiently tested and causes failures.
Change-Id: I623d6fc2cd3ae6fd42d0abab1f8eada465ae57a7
This commit adds the logic for segmentation map initialization and
disable temporal update of segmentation map when error-resilient
mode is on. It fixes the enc/dec mistmates (release build) and
assertions(debug) when both aq-mode and error-resilient are on.
Change-Id: Id2155e8b28962cf1f64494f4df0c8d79499b6890
This commit allows the codec to add motion vector pairs into
the candidate list. It further improves the compression performance
by 0.1% across derf, hevcmr, stdhd, and hevchr sets without adding
encode/decode time.
Change-Id: I88d36da25a2a89bb506d411844af667081eba98b
The '110' prefix on a final byte indicates a superframe marker. Coded
data is not allowed to use this pattern on a final byte.
Code |state - l_base| little endian with the following prefix scheme:
Prefix '00': Single byte coded state.
Prefix '01': Two bytes le coded state.
Prefix '10': Three bytes le coded state.
Change-Id: Ibc953b67675b567394b93de39b7cb22cadc47435
Prior to this patch, read_inter_block_mode_info() would
find the nearmv and nearestmv for all modes. Now it does not
search for ZEROMV modes and breaks out early for NEARMV and
NEWMV modes.
Change-Id: Ifa7b1eaf58bb03b9c7792ea5012fef477527d0fd
This commit re-works the reference motion vector stack process
and make it support extended context set. It unifies reference
motion vector checking process for row and column scan, as well as
for single block scan.
Change-Id: I68c05cde93cf8b0ca2ef4d1523399f405bd0a337
There are flaws in current implementation of VP8 multithreading encoder
and decoder as reported in the following issue:
https://code.google.com/p/chromium/issues/detail?id=158922
Although the data race warnings are harmless, and wouldn't cause real
problems while encoding and decoding videos, it is better to fix the
warnings so that VP8 code could pass the TSan test.
To synchronize the thread-shared data access and maintain the speed
(i.e. decoding speed), use multiple mutexes based on mb_rows to reduce
the number of synchronizations needed, make the reads and writes of
the shared data protected, and reduce the number of mb_col writes by
nsync times.
The decoder speed tests showed < 3% speed loss while using 2 ~ 4
threads.
Change-Id: Ie296defffcd86a693188b668270d811964227882
The nominal tx_type for a given mode is used as a context
to encode the actual tx_type for intra.
Results:
derflr: -0.241% BDRATE
hevcmr: -0.366% BDRATE
Change-Id: Icfe7b0a58d79bc6497a06e3441779afec6e01e21
It allows the codec to account for certain corner cases when
processing inter prediction mode entropy coding.
Change-Id: Ied451f4fff26ba579f6556554b8381ff2ccd0003
This commit enables encoder to avoid 8x4 and 4x8 partitions for
scaled reference frames when libvpx is configured and built with
--enable-better-hw-compatibility
Change-Id: I02ad65c386f5855f4325d72570c49164ed52f413
Move the logic for forcing zero_mode after the
(ref_frame & flag_list) check.
This was causing an memory leak under msan:
https://bugs.chromium.org/p/webrtc/issues/detail?id=5402
Change-Id: Ie9d243369f8ed7c332f46178275945331da4fd85
Under --enable-better-hw-compabibility, this commit adds the asserts
that no mv clamping is applied for scaled references, so when built
with this configure option, decoder will assert if an input bitstream
triggger mv clamping for scaled reference frames.
Change-Id: I786e86a2bbbfb5bc2d2b706a31b0ffa8fe2eb0cb
Various additional changes were made to make the experiment
compatible with misc_fixes.
derflr: +0.979%
hevcmr: +0.865%
Speed-wise with --enable-supertx the encoder is only about 10%
slower than without. Decoding impact is about 30% slowdown.
Note this does not work with ext-tx or var-tx yet. That is
a TODO.
Change-Id: If25af4241a7a9efbd28f58eda3c4f044c7a7ef4b
This commit adds a new configure option:
--enable-better-hw-compatibility
The purpose of the configure option is to provide information on known
hardware decoder implementation bugs, so encoder implementers may
choose to implement their encoders in a way to avoid triggering these
decoder bugs.
The WebM team were made aware of that a number of hardware decoders
have trouble in handling the combination of scaled frame reference
frame and 8x4 or 4x8 partitions. This commit added asserts to vp9
decoder, so when built with above configure option, the decoder can
assert if an input bitstream triggers such decoder bug.
Change-Id: I386204cfa80ed16b50ebde57f886121ed76200bf
Add function to compute skin map for a given block, as its
used in several places (cyclic refresh, noise estimation, and denoising).
Change-Id: Ied622908df43b6927f7fafc6c019d1867f2a24eb
Set initial values for these parameters in the vp9_init_layer_context().
This also fixes an issue in the svc-bypass mode when frame flags are
passed via the vpx_codec_encode().
Change-Id: I0968f04672f8d3d2fe2cea6b8a23f79f80d7a8b1
Otherwise, per-segment lossless might mean that some segments are not
lossless and they could still want to use another mode. The per-block
tx points remain uncoded on blocks where (per the segment id) the Q
value implies lossless.
Change-Id: If210206ab1fe3dd11976797370c77f961f13dfa0
For coding block sizes <=16X16, if the block is determined to be skin,
then always allow for that block to be candidate for refresh. So if that
block happens to be on the boost segment(s), segment won't get reset to 0
and delta-q will be applied.
PSNR/SSIM metrics neutral (little/no change) on RTC clips.
Speed increase small/negligible (< 1%).
Some visual improvement on faces in a few RTC clips.
Change-Id: I6bf0fce6f39d820b491ce05d7c017ad168fce7d6
arm-none-linux-gnueabi- is an anachronism and makes building on native
arm platforms more difficult. further, many distros include alternative
cross compilers, e.g., arm-linux-gnueabihf-, so the choice is best left
up to the user.
Change-Id: Id8aaf820ed112b85db2b8518d0e9d8abee1ad85c
avoids picking up defaults if CROSS is forcibly set empty as in:
$ CROSS= ./configure ...
BUG=1121
Change-Id: I6af91959288dede01efe3e5945698ab249eb6ec3
reduce the register count by 1 to avoid xmm6 and unnecessarily
penalizing the other users of the base macro
Change-Id: I59605c9a41a31c1b74f67ec06a40d1a7f92c4699
In 32-bit build with --enable-shared, there is a lot of
register pressure and register src_strideq is reused.
The code needs to use the stack based version of src_stride,
but this doesn't compile when used in an lea instruction.
This patch also fixes a related segmentation fault caused by the
implementation using src_strideq even though it has been
reused.
This patch also fixes the HBD subpel variance tests that fail
when compiled without disable-optimizations.
These failures were caused by local variables in the assembler
routines colliding with the caller's stack frame.
Change-Id: Ice9d4dafdcbdc6038ad5ee7c1c09a8f06deca362
H/V intra mode was only enabled for bsize < 16x16,
enable it also for bsize=16x16.
Metrics are neutral with this change:
Overall very small gain (0.1%), small visual gain on some RTC clips.
Change-Id: Ib2d7a44382433bfc11cf324aa3cc5c382ea9e088
For the experiment of EXT_REFS, removed the previous special handling
on the new last 3 references, i.e. LAST2_FRAME, LAST3_FRAME, and
LAST4_FRAME, at the decoder, so that these new last references are
treated the same way as the other 3 references (LAST_FRAME,
GOLDEN_FRAME, and ALTREF_FRAME). Encoder changes have been made
accordingly to realize this flexibility.
Change-Id: Ic6546f9443b4377bb7e7b101bfa3e70a8b8d1c65
The DST2 is implemented by input alternate sign-flip, followed
by DCT, followed by output reversal.
Results are roughly the same, but it should be easier to optimize
the DST2.
[Interestingly a mtrix multuiply implementation is about 0.1%
better].
Change-Id: If9ae5fdba87767fb0e6c163a62b77ee66a8d3afc
For testing implemented a fixed pattern and delta, 1 pass,
fixed Q, low delay mode.
This has not in any way been tuned or optimized.
Change-Id: Icf9b57c3bb16cc5c0726d5229009212af36eb6d9
(copied from VP9)
The one pass VBR mode selects a Q range based on a
moving average of recent Q values. This calculation
should have been excluding arf overlay frames as these
are usually coded at the highest allowed value. Their
inclusion skews the average and can cause it to drift
upwards even when the clip as a whole is undershooting.
As such it can undermine correct adaptation of the allowed
Q range especially for easy content.
Change-Id: I9e12da84e12917e836b6e53ca4dfe4f150b9efb1
For testing implemented a fixed pattern and delta, 1 pass,
fixed Q, low delay mode.
This has not in any way been tuned or optimized.
Change-Id: Idf5ee179b277fa15d07a97f14f2ce5bbaae80a04
The one pass VBR mode selects a Q range based on a
moving average of recent Q values. This calculation
should have been excluding arf overlay frames as these
are usually coded at the highest allowed value. Their
inclusion skews the average and can cause it to drift
upwards even when the clip as a whole is undershooting.
As such it can undermine correct adaptation of the allowed
Q range especially for easy content.
Change-Id: I7d10fe4227262376aa2dc2a7aec0f1fd82bf11f9
The culprit is on the decode side xd->lossless[i] setup was in wrong
location where segment features are not yet decoded.
Also on the encoder side, transform mode was not set consistently
between when tx_mode is selected and how tx_mode is enforced in
tx size selection.
Change-Id: I4c4c32188fda7530cadab9b46d4201f33f7ceca3
Keep track of frame indexes for the references, and
constrain inter mode search for reference with same
temporal alignment.
Improves speed by about ~15%, no noticeable loss in
compression performance.
Change-Id: I5c407a8acca921234060c4fcef4afd7d734201c8
Lower the threshold for splitting 32x32->16x16 based on average variance,
and add lower bound condition for this split to occur. This prevents
unneccassry splitting for areas with very low variance.
Change-Id: Ibeb33b3d993632c2019f296eb87ef3b7e3568189
For non-rd variannce partition, speed >= 5:
Adjustments to reduce dragging artifcat of background area near
slow moving boundary.
-Decrease base threshold under low source noise conditions.
-Add condition to split 64x64/32x32 based on average variances
of lower level blocks.
PSNR/SSIM metrics go down ~0.7/0.9% on average on RTC set.
Visually helps to reduce dragging artifact on some rtc clips.
Change-Id: If1f0a1aef1ddacd67464520ca070e167abf82fac
1) Add VP10_XFORM_QUANT_SKIP_QUANT mode for vp10_xform_quant
2) Let encode_block call vp10_xform_quant so that its code flow
is clear
Change-Id: I122d5cf6a089f444ae018f3e4bf844be847e17ee
Reallocate the xmm register usage so that no ARCH_X86_64 required.
Reduce memory access to the left neighbor by half.
Speed up by single digit on big core machine.
Change-Id: I392515ed8e8aeb02e6a717b3966b1ba13f5be990
This commit makes the sub8x8 block rate-distortion optimization
scheme use precise motion compensated prediction to compute the rd
cost. It fixes a potential buffer overflow issue related to sub8x8
motion search on scaled reference frame.
Change-Id: I4274992ef4f54eaacfde60db045e269c13aaa2de
GET_GOT modifies the stack pointer so the offset for left's address will
be wrong if loaded afterword.
Change-Id: Iff9433aec45f5f6fe1a59ed8080c589bad429536
Relocate the function from SSSE3 to SSE2, Unroll loop from 16 to 8,
and reduce mem access to left.
Speed up by single digit in ./test_intra_pred_speed on big core
machines.
Change-Id: I2b7fc95ffc0c42145be2baca4dc77116dff1c960
This commit allows the codec to analyze the reference motion vector
candidate list and adaptively reduce the size of inter prediction
mode set.
Change-Id: Ied6a403843b860d66f26ed485c1825c05c71bdfc
This commit re-works the entropy coding scheme of the motion
compensated prediction modes. It allows more flexible hyperplane
partition for precise classification.
Change-Id: Iba5035c76691946cf1386b6c495e399c3d9c8fc5
Xcode 7 refuses to link to x86 and x86_64 code that's built for
iphone sim, so add an extra command line flag that forces iosbuild
to use darwin15 targets.
Change-Id: I2228d458f5cccf4d26866040380a974f88d9d360
This commit enables the new temporal filter system for VP9. For
speed 1, it improves the compression performance:
derf 0.54%
stdhd 1.62%
Change-Id: I041760044def943e464345223790d4efad70b91e
The 8bit temporal filter has changed to use non-local means.
This fix adds the same change into the high bitdepth code.
Change-Id: I3375b13a7d914fc8aa9eb4aac1d2e4d9b74b782f
This change has been imported from VP9 and
alters the nature and use of exhaustive motion search.
Firstly any exhaustive search is preceded by a normal step search.
The exhaustive search is only carried out if the distortion resulting
from the step search is above a threshold value.
Secondly the simple +/- 64 exhaustive search is replaced by a
multi stage mesh based search where each stage has a range
and step/interval size. Subsequent stages use the best position from
the previous stage as the center of the search but use a reduced range
and interval size.
For example:
stage 1: Range +/- 64 interval 4
stage 2: Range +/- 32 interval 2
stage 3: Range +/- 15 interval 1
This process, especially when it follows on from a normal step
search, has shown itself to be almost as effective as a full range
exhaustive search with step 1 but greatly lowers the computational
complexity such that it can be used in some cases for speeds 0-2.
This patch also removes a double exhaustive search for sub 8x8 blocks
which also contained a bug (the two searches used different distortion
metrics).
For best quality in my test animation sequence this patch has almost
no impact on quality but improves encode speed by more than 5X.
Restricted use in good quality speeds 0-2 yields significant quality gains
on the animation test of 0.2 - 0.5 db with only a small impact on encode
speed. On most natural video clips, however, where the step search
is performing well, the quality gain and speed impact are small.
Change-Id: Iac24152ae239f42a246f39ee5f00fe62d193cb98
4x4 Intra predictor implemented with MMX is replaced with SSE2.
Segfault in change 315561 when decoding vp8 is taken care of.
Change-Id: I083a7cb4eb8982954c20865160f91ebec777ec76
Fix a bug in vp10_has_right;
Some cosmetic changes.
Tiny performance improvement (0.02%~0.04%) on derflr and hevcmr.
Change-Id: Iee829003a20f32d6185a08bab2bd4201806be2b3
Fix copied over from VP9 master to VP10 master.
Do not reset the alt ref active flag when overlaying the middle
arf(s) of a multi arf group.
Change-Id: I1b7392107e7c675640d5ee1624012f39cc374c58
use CONFIG_VP[89] to protect white-box tests and drop redundant
uses of CONFIG_VP9 in variable assignments within that block
Change-Id: Id3c6cf5c7822aa161b19768b295f58829a1c6447
For non-rd variance partition: Adjust variance threhsold based
on noise level estimate. This change allows the adjustment to be
updated more frequently.
Change-Id: Ie2abf63bf3f1ee54d0bc4ff497298801fdb92b0d
Relocate the function from SSSE3 to SSE2, Unroll loop from 8 to 4,
and reduce mem access to left.
Speed up by >20% in ./test_intra_pred_speed.
Change-Id: Ie48229c2e32404706b722442942c84983bda74cc
Relocate the function from SSSE3 to SSE2, Unroll loop from 4 to 2,
and reduce mem access to left.
Speed up by >20% in ./test_intra_pred_speed.
Change-Id: Ib9f1846819783b6e05e2a310c930eb844b2b4d2e
The range_check is not used because the bit range
in fdct# is not correct. Since we are going to merge in a new version
of fdct# from nextgenv2, we won't fix the incorrect bit range now.
Change-Id: I54f27a6507f27bf475af302b4dbedc71c5385118
For low resolutions, whem 4x4downsample is used for variance,
use the same force split (that is used for 8x8downsample) for 16x16 blocks.
No change in metrics. Small improvement visually.
Change-Id: I915b9895902d0b9a41e75d37fee1bf3714d2366d
the loop filter level is transmitted as 6-bits + sign so needs to be clamped in
the delta + absolute case.
BUG=https://bugzilla.mozilla.org/show_bug.cgi?id=1224363
Change-Id: Icbdca4fdbf043466429bd5c9d59dbe913bf153bc
the quantizer is transmitted as 7-bits + sign so needs to be clamped in
the delta + absolute case.
BUG=https://bugzilla.mozilla.org/show_bug.cgi?id=1224361
Change-Id: I9115f5d1d5cf7e0a1d149d79486d9d17de9b9639
1) Add facade to quantize b/fp/dc version so that their interface
are the same.
2) Merge vp10_xform_quant b/fp/dc version to one function so that
the code flow in encodemb.c is clear
Change-Id: Ib62d6215438fc2d07f4e7e72393f964832d6746f
This is so we may update level at any time (e.g., to be used
for setting thresholds in variance-based partition).
Change-Id: I32caad2271b8e03017a531f9ea456a6dbb9d49c7
This commit adds top-right corner and collocated block into the
reference motion vector candidate check list.
Change-Id: I892a4c7fb04ddda44e0f9dfe769471252d40c42b
Estimate angle histogram using gradient analysis, then skip those
angles that are unlikely to be chosen.
On ext-intra experiment, turning off filter-intra modes:
for all-key-frame setting, computation overhead is reduced
by about 40%, coding gain dropped from +2.08% to +1.96% (derflr);
with kf-max-dist=150, computation overhead is reduced
by about 60%, coding gain dropped from +0.58% to +0.49% (derflr).
Change-Id: I36687410fb10561b8e1a8eebb1528cf17755bd5b
Under certain denoising conditons, check for re-evaluation of
zero_last mode if best mode was golden reference.
Change-Id: Ic6cdfd175eef2f7d68606300c7173ab6654b3f6e
Reduce mem access to left. Speed up by 10% in ./test_intra_pred_speed
with the same instruction size.
Change-Id: Ia33689d62476972cc82ebb06b50415aeccc95d15
For non-rd variance partition: only allow minmax computation
(which currently has no arm-neon optimization) for speeds < 8.
Performance loss is small: On RTC set with speed 8, few clips lose ~2/3%,
average loss is < 1%.
Change-Id: Ia9414f4d0b77dc83c3e73ca8de5d903f64b425ce
Relocate h_predictor_4x4 from SSSE3 to SSE2 with XMM registers.
Speed up by ~25% in ./test_intra_pred_speed.
Change-Id: I64e14c13b482a471449be3559bfb0da45cf88d9d
Change initial state of noise level, and only update
denoiser with noise level when estimate is done.
Change-Id: If44090d29949d3e4927e855d88241634cdb395dc
Add inv_txfm and highbd_inv_txfm as facades of inverse transform
such that the code flow in encodemb.c can be simpler
Change-Id: Iea45fd22dd8b173f8eb3919ca6502636f7bcfcf7
Check all motion vectors in the immediate above and left blocks if
the reference conditions matched.
Change-Id: I8bf33bfcee99e8150232c7681fdeade307024272
For denoising, and for noise level above threshold, re-evaluate
ZEROMV for mode selection after denoising.
Current change only does this check if selected best mode (before denoising)
was intra.
Change-Id: I4b1435b68d26c78f7597b995ee7bff0ddd5f9511
This commit ports the motion vector stack from motion field
analyzer to the encoding and decoding pipeline.
Change-Id: Ie283c1e1a15b4c17a1c7c175ce322bf053bb7840
Always round sum error and sum square error toward zero in variance
calculations. This prevents variance from becoming negative.
Avoiding rounding variance at all might be better but would be far
more invasive.
Change-Id: Icf24e0e75ff94952fc026ba6a4d26adf8d373f1c
This commit allows the codec to analyze the motion field in the
avaiable above and left neighboring area to produce a set of
reference motion vectors for each reference frame. These reference
motion vectors are ranked according to the likelihood that it will
be picked.
Change-Id: I82e6cd990a7716848bb7b6f5f2b1829966ff2483
This change makes sure last reference with zero mv
is always checked for mode selection.
No change in metrics.
Change-Id: Iaf01877bf34272b966c78bfe18daad882a0a419e
Reduces the transform optons for INTRA as well as INTER when
transform size is 16x16 to not use any of the DSTs.
Thus, a total of 10 options are used for 16x16, while 4x4
and 8x8 still uses 17 options.
derflr/hevchd actually improves a little, while hevcmr drops
a little.
About 10% speed improvement.
Change-Id: I920a182231e052cdd622f8bb67085c16c572cb1e
These primitive variables are commonly required by many other
experiments as well. The use of n4_w and n4_h was originally
introduced in the vp9 decoder implementation.
Change-Id: I93d701d891e3860f31150031e3b9a2b29a3993d2
Under the experiment of EXT_REFS: LAST2_FRAME, LAST3_FRAME, and
LAST4_FRAME.
Coding efficiency: derflr +1.601%; hevchr +1.895%
Speed: Encoder slowed down by ~75%
Change-Id: Ifeee5f049c2c1f7cb29bc897622ef88897082ecf
the final sum may use up to 26 bits
+ add a unit test
+ disable the sse2 as the result will rollover; this will be fixed in a
future commit
Change-Id: I2a49811dfaa06abfd9fa1e1e65ed7cd68e4c97ce
Change on affects 1 pass CBR.
On key frame, temporal layer_id is reset to 0 for 1 pass CBR,
but since "layer" is reset, the svc.layer_context[layer].is_key_frame
was not correspondingly set properly.
Change-Id: I08f6da0a55ac7429ccfbaddfb7be14479e43543b
tm_predictor_4x4 is implemented with SSE2 using XMM registers.
Speed up by ~25% in ./test_intra_pred_speed.
Change-Id: I25074b78d476a2cb17f81cf654bdfd80df2070e0
--disable-XXX has the effect of disabling all extensions above it, e.g.,
--disable-ssse3 disables ssse3-avx2.
Change-Id: If02b44ca71ee12e4acb12010db8593a7989f2a9d
Remove MISC_FIXES flags except for the changes on MV precision, which
has a 0.1% performance drop.
On derflr, the impact is -0.012%.
Change-Id: I0a74e5a212dd0cb827192a318c92a714c9681e45
Small changes to the best quality default speed trade off.
Some speedup settings are worth while even for best quality as they
have only a very small impact on quality but a significant impact on
encode time.
These changes give as much as a further 50-60% increase in encode
speed for my test animations clip with minimal impact on quality.
For this sequence these changes improve the best quality encode speed
to about the same level as good quality speed 0 in Q3 2015 whilst
retaining the large quality gain of over 1 db
For many natural videos though the quality difference from good 0
to best is much smaller.
Change-Id: I28b3840009d77e129817a78a7c41e29cb03e1132
Reset the effective range of inter_tx_size, instead of the entire
array in the rate-distortion optimization loop.
Change-Id: Id453fbd6dddfe69f4e451ba8518c083326d5dd53
This is simpler than the previous scheme, which tried to allocate
the CRITICAL_SECTION struct in a thread-safe manner before it
could use it to run the wrapped function in a thread-safe manner.
Change-Id: I172e5544e5f16403a3a0e5e2b9104b1292a0d786
This change alters the nature and use of exhaustive motion search.
Firstly any exhaustive search is preceded by a normal step search.
The exhaustive search is only carried out if the distortion resulting
from the step search is above a threshold value.
Secondly the simple +/- 64 exhaustive search is replaced by a
multi stage mesh based search where each stage has a range
and step/interval size. Subsequent stages use the best position from
the previous stage as the center of the search but use a reduced range
and interval size.
For example:
stage 1: Range +/- 64 interval 4
stage 2: Range +/- 32 interval 2
stage 3: Range +/- 15 interval 1
This process, especially when it follows on from a normal step
search, has shown itself to be almost as effective as a full range
exhaustive search with step 1 but greatly lowers the computational
complexity such that it can be used in some cases for speeds 0-2.
This patch also removes a double exhaustive search for sub 8x8 blocks
which also contained a bug (the two searches used different distortion
metrics).
For best quality in my test animation sequence this patch has almost
no impact on quality but improves encode speed by more than 5X.
Restricted use in good quality speeds 0-2 yields significant quality gains
on the animation test of 0.2 - 0.5 db with only a small impact on encode
speed. On most clips though the quality gain and speed impact are small.
Change-Id: Id22967a840e996e1db273f6ac4ff03f4f52d49aa
This commit re-designs the alternate reference frame generation
process. It employs non-local mean approach to produce more stable
pixel estimation for alternate reference frame. It improves the
compression performance gains:
derf 0.5%
hevcmr 0.8%
stdhd 1.3%
hevchr 1.0%
The encoding time at speed 0 is not affected.
Change-Id: Iaa757f0da189ce93812d69617a81bf630d449848
This reverts commit 380a5519cc.
This causes an assertion failure in debug_check_frame_counts() which
probably isn't valid with this change; leaving the investigation for
later now.
Change-Id: Ieda5ca811ed2fa50a0cc6935919a8d10dca996e0
This function now has an AVX intrinsics version which is about 80%
faster compared to the C implementation. This provides a 2-4% total
speed-up for encode, depending on encoding parameters. The function
utilizes 3 properties of the cost function lookup table, constructed
in 'cal_nmvjointsadcost' and 'cal_nmvsadcosts'.
For the joint cost:
- mvjointsadcost[1] == mvjointsadcost[2] == mvjointsadcost[3]
For the component costs:
- For all i: mvsadcost[0][i] == mvsadcost[1][i]
(equal per component cost)
- For all i: mvsadcost[0][i] == mvsadcost[0][-i]
(Cost function is even)
These must hold, otherwise the AVX version of the function cannot be used.
Change-Id: I6c2791d43022822a9e6ab43cd124a773946d0bdc
the return value of enabled, which may be empty, is handled by the for
loop. this avoids making an unnecessarily long command line which may
fail in certain cases.
Change-Id: Ib88ecbbe2c0f6d7debb600b4caed4884497263b1
This commit fixes an encoding failure case triggered when early
termination feature is turned on for transform block size search.
It resolves the corresponding enc/dec mismatch issue.
Change-Id: I2c5b7d8b1efe25fe3810e6ed307f4b1865dede49
Change is only for real-time mode, speed >= 5, and non-screen content mode.
Add bias to zero/low motion for big blocks, if noise estimation
is enabled and noise level is above threshold.
Change-Id: I3a0a4608ede6aa535bda6eca528d20f8aba738e7
For 1 pass CBR mode: increase waiting time after key frame
before we start sampling rate control behavior for determining
resize. This change need to disable one internal resize(DownUp)
temporally since it requires a longer clip to do so.
Change-Id: If21beda1be23f169ee541ab4dd642f718347887a
Use same setting for speed 5 (as it is for speed > 5).
Change is only for real-time (non-rd) mode.
Change-Id: I830250eac654328373cb318baa89d4f0e63942e1
Reduces Linux perf estimated cycle count for pack_mb_tokens on a
lossless encode on my desktop from 61858501855 to 48154040219 or from
26% of the overall profile to 21%.
Change-Id: I9ca3426d7e3272bc7f7030abda4f0d0cec87fb4a
This reverts commit f1342a7b07.
This breaks 32-bit builds:
runtime error: load of misaligned address 0xf72fdd48 for type 'const
__m128i' (vector of 2 'long long' values), which requires 16 byte
alignment
+ _mm_set1_epi64x is incompatible with some versions of visual studio
Change-Id: I6f6fc3c11403344cef78d1c432cdc9147e5c1673
Adds a new interpolation experiment.
Improves entropy coding to send the filter type only if
the motion vectors have subpel components.
Adds one new 8-tap smooth filter, and tweaks the others.
derflr: +0.695%
hevcmr: +0.305%
About 5% encode slowdown. No visible impact for decoding.
Also makes the interpolation framework flexible to support both
strictly interpolating filters as well as non-interpolating
filters that filter integer offsets. This is mainly for
further experimentation and if not found useful the code will
be removed.
Change-Id: I8db9cde56ca916be771fe54a130d608bf10786e6
Add threshold/condition on spatial_variance and brightness level.
Modification to normalization of block variance.
Change resolution limit below which we disable noise estimation.
Change-Id: If5be08a26ceda351242d8a58d2f0bc88c0a918f0
This function now has an AVX intrinsics version which is about 80%
faster compared to the C implementation. This provides a 2-4% total
speed-up for encode, depending on encoding parameters. The function
utilizes 3 properties of the cost function lookup table, constructed
in 'cal_nmvjointsadcost' and 'cal_nmvsadcosts'.
For the joint cost:
- mvjointsadcost[1] == mvjointsadcost[2] == mvjointsadcost[3]
For the component costs:
- For all i: mvsadcost[0][i] == mvsadcost[1][i]
(equal per component cost)
- For all i: mvsadcost[0][i] == mvsadcost[0][-i]
(Cost function is even)
These must hold, otherwise the AVX version of the function cannot be used.
Change-Id: I184055b864c5a2dc37b2d8c5c9012eb801e9daf6
This commit refactors the rate-distortion optimization scheme for
transform block coding. When both ext-tx and var-tx experiments
are turned on, the encoding time for bus_cif at 1000 kbps goes down
from 706377 ms to 666503 ms (5.6% speed-up). The coding statics
remain unchanged.
Change-Id: I20835db573725580aad79c16220f799ce01f2093
When using FLIPADST, the vp10_inv_txfm_add functions used to flip
the destination array, add the result of the inverse transform, to it
and then flip the destination back. This has been replaced by
flipping the result of the inverse transform before adding it to the
destination. Up-Down flipping is done by negating the destination
stride, and staring from the bottom, so it should now be free.
Left-right flipping is done with the usual SSE2 instructions in the
optimized code.
The C functions match the SSE2 functions as expected, so the C functions
now do the flipping as well when required. Adding this cleanly required
some refactoring of the C functions, but there is no measurable
performance impact when ext-tx is not enabled.
Encode speedup with ext-tx enabled is about 3%.
Change-Id: I5b04e5d720f0b9f0d54fd8607a8764f2314c7234
Currently there are two parts in this experiment: extra directional intra
prediction modes and the filter intra modes migrated from the nextgen branch.
Several macros are defined in "blockd.h" to provide controls of the experiment
settings. Setting "DR_ONLY" as 1 (default is 0) means we only use directional
modes, and skip the filter-intra modes; "EXT_INTRA_ANGLES" (default is 128)
defines the number of different angles we want to support; setting
"ANGLE_FAST_SEARCH" as 1 (default is 1) means we use fast sub-optimal search
for the best prediction angle, instead of exhaustive search. The fast search
is about 6 times faster than the exhaustive search, while preserving about
60% of the coding gains.
With extra directional prediction modes (fast search), we observe the following
code gains (number in parentheses is for all-key-frame setting):
derflr +0.42% (+1.79%)
hevclr +0.78% (+2.19%)
hevcmr +1.20% (+3.49%)
stdhd +0.56%
Speed-wise, about 110% slower for key frames, and 30% slower overall.
The gains of filter intra modes mostly add up with the gains of directional
modes. The overall coding gain of this experiment:
derflr +0.94%
hevclr +1.46%
hevcmr +1.94%
stdhd +1.58%
Change-Id: Ida9ad00cdb33aff422d06eb42b4f4e5f25df8a2a
Change is only for real-time mode, speed > 5, and non-screen content mode.
Bias is based on block size and motion vector level (motion above some threshold).
Helps to improves stability in background from lightning changes.
PSNR/SSIM metrics on RTC set almost no change/neutral (within +/- 0.1).
Change-Id: I7eac13c1ae10be4ab1f40acc7f9f1df5653ece9d
Jenkins per-commit test need to be expedited as more experiments are
added into the nextgenv2 branch. This patch does the following:
thread test: change the length of test clip from 5 frames to 3 frames;
only test speed 1.
ArfFreq test: marked as "large".
The tests marked as "large" will be removed from per-commit test
(to nightly test).
Change-Id: I62b373c52b481dcd281e741ebf5098408a97ff4d
Only use non-zero threshold(s) for breakout if
the motion level of the current tested mode is low.
Change-Id: I22aae961cc42371b49d3f648560181cc54708502
This patch eliminates the copying of data when using FLIPADST forward
transforms, by incorporating the necessary data flipping into the
load_buffer_* functions of the SSE2 optimized forward transforms. The
load_buffer_* functions are normally inlined, so the overhead of copying
the data is removed and the overhead of flipping is minimized. Left to
right flipping is still not free, as the columns need to be shuffled in
registers.
To preserve identity between the C and SSE2 implementations, the
appropriate C implementations now also do the data flipping as part of
the transform, rather than relying on the caller for flipping the input.
Overall speedup is about 1.5-2% in encode on my tests. Note that these
are only the forward transforms. Inverse transforms to come in a later
patch.
There are also a few code hygiene changes:
- Fixed some indents of switch statements.
- DCT_DCT transform now always use vp10_fht* functions, which dispatch
to vpx_fdct* for DCT_DCT (some of them used to call vpx_fdct*
directly, some of them used to call vp10_fht*).
Change-Id: I93439257dc5cd104ac6129cfed45af142fb64574
These tables were out of sync with the indexing enum since the
refactoring in commit 4f16f119 (change 303389), due to the removal
of the ext_tx_to_txtype lookup table. This patch just puts them
back in order.
Change-Id: Ieb7d57654f61b99b511d54c9ba09abbd5e8d0d14
This commit re-works the rate-distortion optimization scheme for
transform coding. It improves the overall compression performance.
For derf set, the ext-tx experiment provides 2.27% coding gains,
and the new scheme that integrates multiple transform type selection
and recursive transform block partitioning provides a total of 3.24%
coding gains.
Change-Id: Ia1887c4c44b73dfb915d091d96660a99f09d5cc3
This commit hooks up the rate-distortion optimization system to
fully exploit recursive transform block partition and multiple
transform type. The compression performance of the two experiments
largely adds up. For derf set, ext-tx provides additional 2.1%
coding gains on top of the gains due to recursive transform block
partition (0.69%).
Change-Id: I1091fb9545f74e489a6a2489dc3c12f5abd05043
Correctly compute the block size in bit-stream coefficient token
packing. This fixes an enc/dec mismatch at very high bit-rates.
Change-Id: I37bf084731dc660df0c695cad406ddcd0f9eb904
Source noise level estimate is also useful for
setting variance encoder parameters (variance thresholds,
qp-delta, mode selection, etc), so allow it to be used also
if denoising is not on.
Change-Id: I4fe23d47607b4e17a35287057f489c29114beed1
Change-Id: I38952cd55b91f35e5db45bc8e6a20ef25069c464
--ext-refs: extended references - for multi-ref in nextgen
--ext-inter: extended inter - for new_inter/copy_mode in nextgen
--ext-interp: for new interpolation
this avoids redefining vpx_codec_vp9_dx, vpx_codec_vp9_dx_algo in
vp9_encoder_parms_get_to_decoder.cc
Change-Id: I3b89e7a62497227ee32419f1a7d30e4c10a13c05
(cherry picked from commit ca163b85bb)
This commit allows the loop filter to account for the recursive
transform block partition when selecting the filter and mask.
Change-Id: I62b6c2dcc0497cbe1f264b03c46163f55d2c9752
This commit refactors the loop filter selection process to support
variable transform block sizes based filter mask. It disables the
multi-thread loop filter implementation to simplify the experiments.
The speed impact on speed 0 encoding is negligible.
Change-Id: Ia470b6da9ad833fe6eb72d2cbeda9296b21910ec
If a block has all coefficients quantized to zero, the codec will
assume that it uses largest transform block size.
Change-Id: Icd4e8e7cdc4b6af6974f87169e50b040ebfe9020
this avoids redefining vpx_codec_vp9_dx, vpx_codec_vp9_dx_algo in
vp9_encoder_parms_get_to_decoder.cc
Change-Id: I3b89e7a62497227ee32419f1a7d30e4c10a13c05
The old workaround "p = 0 ? 0 : p -1" is misleading.
?: happens before =
assigning back to p truncates to one byte.
Therefore it is equivalent to (p - 1) & 0xFF, but the check just exists
to work around a first pass bug, so let's make the work around more
clear.
https://bugs.chromium.org/p/webm/issues/detail?id=1089
Change-Id: I587c44dd61c1f3767543c0126376f881889935af
Width and height of downscaling resolution should not be lower
than min_width and min_height which can be set as needed, both
are 180 for now.
Change-Id: I34d06704ea51affbdd814246e22ee8d41d991f00
Allows inter and intra tx_types to have different sets of
transforms for different tx_size/sb_type combinations.
Change-Id: Ic0ac1daef7a9fb15c4210271e4d04cd36e5cec8e
This reverts commit 7f56cb2978.
It causes uninitialized reads in the first pass setting up later cost tables.
Change-Id: I2df498df3f5c03eff359f79edf045aed0c618dc9
Rework the rate distortion optimization pipeline. Use precise
distortion metric that accounts for the forward and inverse
transform rounding effect.
Change-Id: Ibe19ce9791ec3547739294cc3012dd9e11f4ea49
This commit makes the coefficient token packtization process account
for variable transform block sizes supported in a single processing
block. It fixes an enc/dec mismatch issue when var-tx, ext-tx, and
misc-fixes experiments are all turned on.
Change-Id: I2e8946e6f72de567603a568debbadad11196430c
Properly update the transform type counts in key frame coding at
decoder. It fixes an enc/dec mismatch issue when both ext-tx and
misc-fixes are turned on.
Change-Id: I1e40a77c8d8157d5ff254b072ce474d8dfbaa3ae
Remove delta index 254 from probability remapping and subexp coding.
Saves 1-bit when the delta index is 129.
Change-Id: I88aba565fc766b1769165be458d2efd3ce45817e
Adjust variance threshold, delta-qp, and intra penalty cost,
based on estimated noise level in source.
Replace denoising_on with a level value=L/M/H.
Change-Id: I0c017dae75a5d897367d2c42dec26f2f37e447c1
EXT_TX introduces some new symbols to be decoded.
The encoder counts how many times these are used.
In multithreaded mode, the counts from the worker threads
need to be accumulated into the main thread.
This change means that VP10/VPxEncoderThreadTest now works
with more choices of cpu-used and number of passes.
Change-Id: Ibe7e6a3c58145265f4ead155ff98fb4cb37c3513
Properly reset the early termination flag in the recursive transform
block partitioning rate-distortion optimization scheme.
Change-Id: Ibfe918f21f11dcb1ec267c09f954c635305cc95a
The option exists specifically to allow for configurations
where the build environment is different from the configure
environment.
Change-Id: I95196fa3c49700251d10ff5d256dc7380e39d0c4
Use inter_block_yrd as rate-distortion optimization for lossless
coding. This fixes transform coefficient buffer swap use case and
resolves the unit test failure related to lossless coding.
Change-Id: I1512dab5ed5760c31f7de21a06e8d9ed1eb081fa
The old workaround "p = 0 ? 0 : p -1" is misleading.
?: happens before =
assigning back to p truncates to one byte.
Therefore it is equivalent to (p - 1) & 0xFF, but the check just exists
to work around a first pass bug, so let's make the work around more
clear.
https://code.google.com/p/webm/issues/detail?id=1089
Change-Id: Ia6dcc8922e1acbac0eeca23a4d564a355c489572
This commit properly resets the recursive transform block partition
array in the settings of using largest transform block size at frame
header level. It fixes one of the unit test failure related to the use
of frame level fixed transform block size with 440 color format.
Change-Id: I6750f323e2c2510c080ffc3af82ce2041f4f60b8
The custom LCG is based on the POSIX recommend constants for a 16-bit
rand(). This implementation uses less computation than typical standard
library procedures which have been extended for 32-bit support, is
guaranteed to be reentrant, and identical everywhere.
Change-Id: I3140bbd566f44ab820d131c584a5d4ec6134c5a0
Ref: http://pubs.opengroup.org/onlinepubs/9699919799/functions/rand.html
Bug relating to issue:- http://b/25090786
base_frame_target is supposed to track the idealized bit
allocation based on error score and not the actual bits
allocated to each frame.
The clamping of this value based on the VBR min and max pct values
was causing a bug where in some cases the loop that adjusts the
active max quantizer for each GF group was running out of bits at
the end of a KF group. This caused a spike in Q and some ugly artifacts.
A second change makes sure that the calculation of the active
Q range for a group DOES, however, take account of clamping.
Change-Id: I31035e97d18853530b0874b433c1da7703f607d1
This commit makes the recursive transform block partitioning properly
handle the non-420 color format. It resolves an enc/dec mismatch
issue in that setting when var-tx experiment is turned on.
Change-Id: I48a91de02c11b3153f897d1cca0ae948eec15605
This commit fixes an enc/dec mismatch issue in recursive transform
partitioning experiment due to merge conflict.
Change-Id: I66146ef806c008902c91d54f4f8c7ccf47996b78
Periodically estiamte noise level in source, and only denoise
if estimated noise level is above threshold.
Change-Id: I54f967b3003b0c14d0b1d3dc83cb82ce8cc2d381
Add the row and column index to the argument list of unit functions
called by foreach_transformed_block wrapper. This avoids the
repeated internal parsing according to the block index.
Change-Id: Ie7508acdac0b498487564639bc5cc6378a8a0df7
This commit fixes the merge conflicts between master and nextgenv2 and
disable early termination in choose_tx_size() to avoid failure in test.
The test failures are pre-existing, some of the issue were fixed in
masterbase already, so will have another merge to introduce the fixes.
Change-Id: Ib71889661955e73aedbb4db49d8be70425281dcb
Temporarily reset the transform type in the inter modes when
recursive transform block partitioning is used. This resolves an
enc/dec mismatch issue in nextgenv2 codebase when both var-tx and
ext-tx experiments are turned on.
Change-Id: I2543f0a567243da95b237752d46964b07b669ad9
Clear the compiler errors when both high bit-depth and recursive
transform block partition experiments are enabled.
Change-Id: If0b6396851f10c28b4f26350322ccd1ba2fc9aff
A new version of vp9_highbd_error_8bit is now available which is
optimized with AVX assembly. AVX itself does not buy us too much, but
the non-destructive 3 operand format encoding of the 128bit SSEn integer
instructions helps to eliminate move instructions. The Sandy Bridge
micro-architecture cannot eliminate move instructions in the processor
front end, so AVX will help on these machines.
Further 2 optimizations are applied:
1. The common case of computing block error on 4x4 blocks is optimized
as a special case.
2. All arithmetic is speculatively done on 32 bits only. At the end of
the loop, the code detects if overflow might have happened and if so,
the whole computation is re-executed using higher precision arithmetic.
This case however is extremely rare in real use, so we can achieve a
large net gain here.
The optimizations rely on the fact that the coefficients are in the
range [-(2^15-1), 2^15-1], and that the quantized coefficients always
have the same sign as the input coefficients (in the worst case they are
0). These are the same assumptions that the old SSE2 assembly code for
the non high bitdepth configuration relied on. The unit tests have been
updated to take this constraint into consideration when generating test
input data.
Change-Id: I57d9888a74715e7145a5d9987d67891ef68f39b7
This causes the output of find_ref_mvs() to always be unique or zero.
A nice side-effect of this is that it also causes the output of
find_ref_mvs_sub8x8() to be unique-or-zero, and it will not ignore
available candidate MVs under certain conditions.
See issue 1012.
Change-Id: If4792789cb7885dbc9db420001d95f9b91b63bfa
Added optimization of the 8 bit assembly quantizer routines. This makes
these functions up to 100% faster, depending on encoding parameters.
This patch maskes the encoder faster in both the high bitdepth and 8bit
configurations. In the high bitdepth configuration, it effects profile 0
only.
Based on my profiling using 1080p input the net gain is between 1-3% for
the 8 bit config, and around 2.5-4.5% for the high bitdepth config,
depending on target bitrate. The difference between the 8 bit and high
bitdepth configurations for the same encoder run is reduced by 1% in all
cases I have profiled.
Change-Id: I86714a6b7364da20cd468cd784247009663a5140
VP8E_UPD_ENTROPY, VP8E_UPD_REFERENCE and VP8E_USE_REFERENCE have been
deprecated since the initial public release
Change-Id: Ied16b441eec13434d85f1ab115d49ccaf5f2f7b0
Some more testing of this patch would probably be useful, but I
think the basics of it should work fine now.
See issue 1035.
Change-Id: I4a36d58f671c5391cb09d564581784a00ed26245
This experiment allows using full above/right edges for all transform
sizes whenever available (for d45/d63), and adds bottom/left edges for
d207.
See issue 1043.
Change-Id: I5cf7f345e783e8539bb6b6d2c9972fb1d6d0a78b
In VP9, the ref MV had to point to a block that itself fully resided
within the visible image, i.e. all borders of the image had to be
within the visible borders of the coded frame. This is somewhat
illogical, and had obscure side effects, e.g. clamping of fairly
reasonable motion vectors such as 0,0 were clipped to negative values
if the block was overhanging on frame edges (such as the last rows
on 1080p content), which makes no sense whatsoever.
Instead, relax clamping constraints such that the ref MVs are allowed
to point to blocks exactly outside the visible edges in both Y as well
as UV planes, including the 8tap filter edges (that's why the offset is
8 pixels + block size).
See issue 1037.
Change-Id: I2683eb2a18b24955e4dcce36c2940aa2ba3a1061
This has various benefits:
- simplify implementations because we don't have to switch between
multiple probability tables depending on frametype
- allows fw subexp and bw adaptivity for partitions/uvmode in keyframes
See issue 1040 point 5.
Change-Id: Ia566aa2863252d130cee9deedcf123bb2a0d3765
Locate them (code-wise) in frame_context, and have them be updated
as any other probability using the subexp forward and adaptive bw
updates.
See issue 1040 point 1.
TODOs:
- real-world default probabilities
- why is counts sometimes NULL in the decoder? Does that mean bw
adaptivity updates only work on some frames? (I haven't looked
very closely yet, maybe this is a red herring.)
Change-Id: I23b57b4e5e7574b75f16eb64823b29c22fbab42e
Account for rounding in distortion calculation in k-means;
carry out rounding before duplicates removal of base colors;
replace numbers with macros;
use prefix increment.
Slight coding gain (<0.1%) on screen_content testset.
Change-Id: Ie8bd241266da6b82c7b2874befc3a0c72b4fcd8c
Adjust the qp threshold and consec_zeromv threshold for
limiting cyclic refresh. Also increase the refresh period
when the limit amount is significant, and some code-cleanup.
Small gain in PSNR/SSIM metrics: ~0.25/0.3 gain on RTC set, speed 7.
Change only affects non-screen content.
Change-Id: I1ced87a89a132684c071e722616e445b2d18236a
Adjust the qp threshold based on the denoising setting; not allow
to scale directly from original resolution to one half and vise versa.
Change-Id: I032a9b22f8e1c88de6bb81cf8351367223a3e40d
For the re-encoding (at max-qp) on the detected high-content change:
update rate correction factor, reset rate over/under-shoot flags,
and update/reset the rate control for layered coding.
Change-Id: I5dc72bb235427344dc87b5235f2b0f31704a034a
Changes to the breakout behavior for partition selection.
The biggest impact is on speed 0 where encode speed in
some cases more than doubles with typically less than 1%
impact on quality.
Speed 0 encode speed impact examples
Animation test clip: +128%
Park Joy: +59%
Old town Cross: + 109%
Change-Id: I222720657e56cede1b2a5539096f788ffb2df3a1
This commit re-designs the recursive transform block partition
rate-distortion optimization framework. It allows the encoder to
improve speed by 10%.
Change-Id: I6dd3a7dd428a530d8012e5c6ddc40e650c8b392b
This change (in a new config experiment: universal_hp) removes the
bitstream parsing dependency of the HP MV bit on the ref MV to be
coded. It also cleans up clearing of the HP bit in near/nearestMV,
since HP is always on if it's set in the frame header.
This admittedly doesn't clean up the crap that could be cleaned up,
but that's mostly because I think this needs some careful review;
not so much for coding style, but more from hardware people and from
the codec team on what we/you want. It would also be nice to get some
actual numbers on the real quality impact of this change. If, for
example, hardware people come up and tell us they don't actually care
anymore, we should probably just this code as-is and do nothing (i.e.
discard this patch).
See issue 1036.
Change-Id: Ic9b106f34422aa0f79de0c28125b72d566bd511a
This actually has no effect whatsoever, since the input MVs themselves
are clamped by clamp_mv_ref() already, which is significantly more
restrictive in its bounds.
Change-Id: I4a3a7b2b121ee422c56428c2a12d930c3813c06e
We only write EOSB tokens if we write tokens (i.e. not for skip blocks),
and we write EOSB tokens per-plane instead of per block.
Change-Id: I8d7ee99f8ec50eb7ae809f9f9282c1c91dbf6537
Add palette mode for keyframe luma channel. Palette mode is enabled
when using "--tune-content=screen" in encoding config parameters.
on screen_content testset: +6.89%
on derlr : +0.00%
Design doc (WIP):
https://goo.gl/lD4yJw
Change-Id: Ib368b216bfd3ea21c6c27436934ad87afdaa6f88
This commit makes the rate-distortion optimization for chroma
component support the recursive transform block coding scheme.
Change-Id: I1bfed6d05b0ebb3905cb625222401e2ccbae10f3
If high bit depth configuration is enabled, but encoding in profile 0,
the code now falls back on optimized SSE2 assembler to compute the
block errors, similar to when high bit depth is not enabled.
Change-Id: I471d1494e541de61a4008f852dbc0d548856484f
This commit makes the transform, quantization, tokenization and
their corresponding inverse operations support recursive transform
block coding process.
Change-Id: I71f2ef3a7c2d3db7cfc63c1fd3f1337e8e0360b5
This commit re-designs the bitstream syntax to support recursive
transform block partition. The initial stage targets the inter
prediction residuals.
Change-Id: I556ab3c68c198387a2fd2d02e2b475e83cd417c3
Add the row and column index to the argument list of unit functions
called by foreach_transformed_block wrapper. This avoids the
repeated internal parsing according to the block index.
Change-Id: I42b3578eac258ebaba7a7c74f684de9abab521a6
The artifact occurs periodically when VP9 denoiser is on and
refresh_golden_frame happen. When refresh_golden_frame happen,
we should copy the frame buffer instead of swapping the pointers.
Change-Id: Ib3204c4b04db28ecf439c6d9e61f3d146f04196d
this reduces the number of synchronizations in decode_tiles_mt() and
improves overall performance when the number of threads is less than the
number of tiles
Change-Id: Iaee6082673dc187ffe0e3d91a701d1e470c62924
Small code cleanup. consec_zeromv refresh threshold
does not need to be computed for every super-block.
No change in behavior.
Change-Id: I8c4b1b28072f42b01d917fff6d1f62722f1e1554
The serial decode check is too strict for tile-threaded decoding as
there is no guarantee on the decode order nor which specific error
will take precedence. Currently a tile-level error is not forwarded so
the frame will simply be marked corrupt.
Change-Id: I51cf1e39e44bedeac93746154b36a4ccb2f059b1
Use the existing VP9_SET_SVC control to set the
first spatial layer to encode.
Since we loop over all spatial layers inside the encoder, the
setting of spatial_layer_id via VP9_SET_SVC has no relevance.
Use it instead to set the first_spatial_layer_to_encode,
which allows an application to skip encoding lower layer(s).
Change only affects the 1 pass CBR SVC.
Change-Id: I5d63ab713c3e250fdf42c637f38d5ec8f60cd1fb
When configured with high bit detpth enabled, the 8bit quantize
function stopped using optimised code. This made 8bit content
decode slowly. This commit re-enables the SSSE3 optimisations.
Change-Id: I194b505dd3f4c494e5c5e53e020f5d94534b16b5
We have historically added new bits to cat6 whenever we added a new
transform size (or bitdepth, for that matter). However, we have
always coded these new bits regardless of the actual transform size,
which means that for smaller transforms, we code bits that cannot
possibly be set. The coding (quality) impact of this is negligible,
but the bigger issue is that this allows creating bitstreams with
coefficient values that are nonsensible and can cause int overflows,
which then de facto become part of the bitstream spec. By not coding
these bits, we remove this possibility.
See issue 1065.
Change-Id: Ib3186eca2df6a7a15ddc60c8b55af182aadd964d
This is identical to what the tile size does for the last tile. See
issue 1042 (which covers generalizing the superframe/tile concepts).
Change-Id: I1f187d2e3b984e424e3b6d79201b8723069e1a50
See issue 1051. 6 bits is fairly arbitrary but at least allows writing
delta Q values that are fairly normal in other codecs. I can extend to
8 if people want full range, although I personally don't have any need
for that.
Change-Id: I0a5a7c3d9b8eb3de4418430ab0e925d4a08cd7a0
The resolution check fixs the issue which resets resize_pending
unnecessarily and causes not-bitexact with previous one-step version.
Change-Id: I4e7660b3c8f34f59781e2e61ca30d61080c322de
When configured with high bit detpth enabled, the 8bit quantize
function stopped using optimised code. This made 8bit content
decode slowly. This commit re-enables the SSE2 optimisation
(but not the SSSE3 optimisation).
Change-Id: Id015fe3c1c44580a4bff3f4bd985170f2806a9d9
Temporary fix to denoiser when dynamic resizing is on.
-Reallocate denoiser buffers on resized frame.
-Force golden update on resized frame.
-Don't denoise resized frame, and copy source into denoised buffers.
Change-Id: Ife7638173b76a1c49eac7da4f2a30c9c1f4e2000
For screen-content mode, with frame dropper off, put a limit
on how low encoder buffer can go.
Under hard slide changes, the buffer level can go too low and then
take long time to come back up (in particular when frame-dropping
is not used), which will affect the active_worst and target frame size.
Change-Id: Ie9fca097e05cd71141f978ec687f852daf9de332
Dynamic resizing now support two-steps scaling: first go down to
3/4 and then 1/2. This feature is under a flag which controls the
switch between two-steps scaling and one-step scaling (1/2 only).
Change-Id: I3a6c1d3d5668cf8e016a0a02aeca737565604a0f
This is more a proof of concept than anything else. The problem here
isn't so much how to code it, but rather where to place the resulting
code. All intrapred DSP code lives in vpx_dsp, so do we want the vp10
specific intra pred functions to live there, or in vp10/?
See issue 1015.
Change-Id: I675f7badcc8e18fd99a9553910ecf3ddf81f0a05
The x86 simd expects this. Identical alignment can be found in vp9
and vp10 also. Fixes crashes on 32bit x86 systems.
Change-Id: I229c88d8f696acbef5337c8fa9503528df4e1c40
I've added a few new functions (d45e, d63e, he, ve) to cover the
filtered h/v 4x4 predictors that are vp8-specific, the "correct"
d45 with the correctly filtered bottom-right pixel (as opposed to
the unfiltered version in vp9), and the "broken" d63 with weirdly
filtered bottom-right pixels (which is correctly filtered in vp9).
There may be a minor performance impact on all systems because we
have to do an extra copy of the Above pixel array to incorporate
the topleft pixel in the same array (thus fitting the vpx_dsp API).
In addition, armv6 will have a more serious performance impact b/c
I removed the armv6/vp8-specific assembly. I'm not sure anyone
cares...
Change-Id: I7f9e5ebee11d8e21aca2cd517a69eefc181b2e86
define NOMINMAX to allow the std:: versions to be used; min/max will be
defined transitively via windows.h otherwise
Change-Id: I692b03fa3e70b7a53962d3fd209498f70f712fed
Adds an early termination to the ext_tx search, and also
implements the DST transforms more efficiently.
About 4 times faster with the ext-tx experiment.
There is a 0.09% drop in performance on derflr from 1.735% to
1.648%, but worth it with the speedup achieved.
Change-Id: I2ede9d69c557f25e0a76cd5d701cc0e36e825c7c
Resolved Conflicts in the following files:
configure
vp10/common/idct.c
vp10/encoder/dct.c
vp10/encoder/encodemb.c
vp10/encoder/rdopt.c
Change-Id: I4cb3986b0b80de65c722ca29d53a0a57f5a94316
vp9_filter_block_plane_ss11() and vp9_filter_block_plane_non420()
are only called for the uv planes.
Change-Id: Iacd3b3242c8ce581edd37c8f06d95efc8a0f88a3
The loopfilter masks are now built in the decode loop.
This is done so we can eventually reduce the number of
MODE_INFO structs required by the decoder.
The encoder builds the masks for the entire frame prior
to calling the loopfilter.
Change-Id: Ia2146b07e0acb8c50203e586dfae0c4c5b316f11
When configured with high bitdepth enabled, the 8bit transform
stopped using optimised code. This made 8bit content decode slowly.
Change-Id: I67d91f9b212921d5320f949fc0a0d3f32f90c0ea
Update rate correction factor when we drop the frame due to overshoot.
Only affects when the drop_overshoot feature is on: screen_content_mode = 2.
Change-Id: I67e24de979b4c74744151d2ceb3cd75fec2a1e7a
In the decoder, map this to the output variable vpx_image_t.r_w/h.
This is intended as an improved version of VP9D_GET_DISPLAY_SIZE,
which doesn't work with parallel frame decoding. In the encoder,
map this to a codec control func (VP9E_SET_RENDER_SIZE) that takes
a w/h pair argument in a int[2] (identical to VP9D_GET_DISPLAY_SIZE).
Also add render_size to the encoder_param_get_to_decoder unit test.
See issue 1030.
Change-Id: I12124c13602d832bf4c44090db08c1009c94c7e8
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.
Change-Id: Iab8d2eae96492b71c4ea60c4bce8121cb2a1fe2d
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).
See issue 1016.
Change-Id: Id53b501cbff9323a8239ed4775ae01fe91874b7e
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.
Change-Id: I234a1128d017d284cf767fdb58ef6c59d809f679
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.
BUG=https://code.google.com/p/webm/issues/detail?id=1075
Change-Id: I38c997a0225e53c5dd1b4ddf7935d21362953f76
Always add IOS_VERSION_MIN to darwin arm cflags. The warning occured
because the default (9.0) does not match the value set by configure
(6.0).
BUG=https://code.google.com/p/webm/issues/detail?id=1075
Change-Id: Ia9085ceeca10e057f9eb781c14f07581bb6280a5
- Use the iphoneos SDK path (instead of macosx).
- Detect iOS SDK major version and disable media (armv6) when using
iOS SDK version 9 or higher.
BUG=https://code.google.com/p/webm/issues/detail?id=1075
Change-Id: I12f77dbeee4c0084e8322f6841813da8b5e91c16
these have been supported in tile-threaded decoding since:
b3b7645 vp9_dthread: remove frame_parallel_decoding_mode requirement
Change-Id: Ia5a752db9be937153cf4830d9258752136356d1b
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-Id: I68e2db87875343b3fb9bb407a7709f0088f84072
Consider tha case in which skipping transform coefficients is more
efficient.
derflr +0.13%
hevclr +0.11%
hevcmr +0.14%
hevchr +0.22%
with ext-tx, the impact is -0.02%.
Change-Id: I0aa2965cf9e152396623c2fee62545bd3a3a7f07
remove static from fdct4/8/16/32 in vp10/encoder/dct.c
add prefix vp10_ to fdct4/8/16/32
add vp10/encoder/dct.h
Change-Id: I644827a191c1a7761850ec0b1da705638b618c66
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.
BUG=1074
Change-Id: Ie26ed816a57fa871ba27a72db9805baaaeaba9f3
Reference frame masking logic may skip checking zeromv-last mode.
Fix to avoid this and make sure zero-last is always checked.
No noticeable change in speed, and PSNR/SSIM metrics on RTC set overall
neutral (very small gain ~0.02).
Small visual improvement on few RTC clips.
Change-Id: I26eacdc449126424001a4a64e5ac31949f064417
the range check in dct.c (abs(input[i]) < (1 << bit)) will fail in many
cases. this was broken at the time this check was added
BUG=1076
Change-Id: I3df8c7a555e95567d73ac16acda997096ab8d6e2
the range check in dct.c (abs(input[i]) < (1 << bit)) will fail in the
25-29 range. this was broken at the time this check was added
Change-Id: I8ca9607f6cbdc8be7f47696ffeabbab3ac5727e2
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.
Change-Id: I2ddda094361bf0ac77f8d2ae542e4dc7b2cab158
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
Change-Id: I9c768c0b2047481d1d3b11c2e16b2f8de6eb0d80
This commit removes mbmi_ext_base pointer from MACROBLOCK struct.
Its use case can be fully covered by cpi->mbmi_ext_base pointer.
Change-Id: I155351609336cf5b6145ed13c21b105052727f30
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.
Change-Id: I05f941897cae13fb9275b939d11f93941cb73bee
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.
See issue 1013.
Change-Id: If39c890cad20dff96635720d8c75b910cafac495
In vp9, the bottom MV would be the average of the topright and
bottomleft luma MV (instead of the bottomleft/bottomright luma MV).
See issue 993.
Change-Id: Ic91c0b195950e7b32fc26c84c04788a09321e391
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).
See issue 662.
Change-Id: I23e607c6de150b9f11d1372fb2868b813c322d37
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.
See issue 1039.
Change-Id: I6a8182cc88a16842fb431688c38f6b52d7f24ead
Extend the ext_tx experiment to make the UV inter blocks use
the same transform type as the extended transform type used
for Y.
derflr: +1.792% (about +0.06)
Change-Id: I4a77e1f7764b2e8b523e28f42ba13559dde4f0ca
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.
See issue 1014.
Change-Id: I8629e6abe76f7c1d649f28cd6fe22a675ce4a15d
In decoder, export (eventually) into vpx_image_t.range field. In
encoder, use oxcf->color_range to set it (same way as for
color_space).
See issue 1059.
Change-Id: Ieabbb2a785fa58cc4044bd54eee66f328f3906ce
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.
Change-Id: I7ad8c9a4c8288387d897dd6bdda592f142d8870c
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.
Change-Id: Ic347be60972fa87f7d68310da2a055679788929d
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.
Change-Id: Ib9c3b5ba3cb7851bfb8c37d4f911664bef38e165
Fixes temporal scalability. Updates were inadvertently turned
off for two pass svc causing crashes due to gf_group.index
growing unchecked.
Change-Id: Iff759946bf61bbde70630347cc8fa4d51a8c2d2f
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.
See issue 1060.
Change-Id: I19a3adda4a8662a05f08a9e58d7e56ff979be11e
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.
Change-Id: I2b71d9ff18b3c7499b058d1325a9554de993dd52
Unify the style of fdct4() fdct8() fdct16()
Add fdct32()
Add range_check() at each stage
Add unit test at ../../test/vp10_dct_test.cc
Change-Id: I13f76d9046c3ea473c82024b09a5bc8662e2c28e
Upstream hash: 476366249e1fda7710a389cd41c57db42305e0d4
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
Change-Id: Ie0375e564fc74b3b296744d0039830d2f77b83b6
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.
Change-Id: I37f0177e5759f54e2e6cc6217023d5681de92438
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.
See issue 1030.
Change-Id: If450c74162d35ca63a9d279beaa53ff9cdd6612b
1) copy following files from vpx_dsp/ to vp10/common/
vp10_inv_txfm.c
vp10_inv_txfm.h
vp10_inv_txfm_sse2.c
vp10_inv_txfm_sse2.h
2) change the function prefix "vpx_" to "vp10_" in above files
3) add unit test at vp10_inv_txfm_test.cc
Change-Id: I206f10f60c8b27d872c84b7482c3bb1d1cb4b913
This condition is not effectively in use. The actual reference
frame masking is done in other route.
Change-Id: Ia59c843bcac7243dada92f0f67658d7ce43df5e8
remove 'u' and specify all objects to allow objects with the same
basename to be added and a incremental rebuild to succeed
fixes issue #1067
Change-Id: Id0ebc89be826a026f1bbf21b4e32a2b1af45154d
Take out speed features that affect the compression performance
to simplify the coding route. This commit removes the motion field
mode search used in speed 3.
Change-Id: Ifdf6862cb1ece8261125a56d9d89bcef60758c00
Creates new hybrid transforms combining symmetric DST with
ADST and DCT. Thus a total of 16 transforms are supported.
derfl: +1.659% (up about 0.2%)
Change-Id: Idde1cecdb59527890bf05da740099c3f6a5b9764
WebM files could have CodecId missing in the track headers. Treat those files as
unknown input file type in vpxdec.
Fixes issue #1064.
Change-Id: I6c3bb7b4bd3a4f5c244312482a5996f8b68db3f3
avoid duplicating internal structures and include vp9_dx_iface.c
directly. these had fallen out of sync after the frame-parallel branch
merge.
Change-Id: I604cfbffa95abe2a1c8e906a696f32436b1422ed
this file needs to be reworked to remove the duplication of codec
internals + allow for divergence of vp9 and vp10
Change-Id: I6266b94ccfbc24dae30148f134804b52aa411b88
prevents an int -> vpx_img_fmt_t conversion warning with high-bitdepth
as it modifies the image format
Change-Id: Ie3135d031565312613a036a1e6937abb59760a7e
Access scaled reference frame in the sub8x8 rate-distortion
optimization loop only when the current test mode is an inter mode.
This prevents an ioc warning triggered by sending intra_frame index
to fetch scaled reference frame.
Change-Id: I6177ecc946651dd86c7ce362e3f65c4074444604
For 3 temporal layers, reduce somewhat the
cyclic_refresh_mode_max_mbs_perframe parameter, from 20% to ~14%.
Small increase in PSNR/SSIM metrics.
Change-Id: Ia216fa5474048f1ef7fe3db88cd60dfef2a1bf8a
Does not include DST1 yet.
derflr: +1.437 (8-bit internal), +7.243 (12-bit internal)
with --enable-ext-tx
Change-Id: I91f1759fd2de794755eb6384cda52e80e979cb7d
Change settings for 1 pass CBR.
And only use SET_SVC control(s) if there is at least 1 layer (spatial or temporal).
This allows sample encoder to also work for 1 layer case.
Change-Id: I5b0a33c25afb2f24a3a8aa4ec8ade9afc87cd702
This commit allows the encoder to include sub8x8 inter mode with
scaled reference frame in the rate-distortion optimization scheme.
Change-Id: Ibbe9678801592826ef22566566dcdeeb008350d5
Sync the encoder's buffer offset calculation for sub8x8 block motion
compensated prediction with scaled reference frame to match the
decoder's behavior. This resolves an enc/dec mismatch issue when
sub8x8 inter mode with scaled is turned on.
Change-Id: I4bab3672b007a5ae0c992f8a701341892d2458b0
The fields are always coded in the frame itself, so there is never any
dependency on past frames. In practice, this fixes sign_bias being
ignored when error_resilient_mode=1.
See issue 1011.
Change-Id: I9d134ef6b445ced4d100fa735ce579855a0fa5af
the check performed within the while was redundant; simply place the
accumulation after all tiles are decoded.
Change-Id: I6a74e87257c775fd8bfc8ac4511e4a6ad8f18346
This is based on the original patch optimized for 32bit
platforms by Tamar/Ilya and now uses the x86inc style asm.
The assembly was also modified to support 64bit platforms.
Change-Id: Ice12f249bbbc162a7427e3d23fbf0cbe4135aff2
These frame types cannot make bitstream parsing depend on previous
frames, so the hypothetical combinations of e.g. keyframe=1 and
update_map=0 or keyframe=1 and temporal_update=1 are non-sensical.
Therefore, make it impossible to code such combinations in the vp10
bitstream header.
See issue 1044.
Change-Id: I3f0a83d5c7e3989541a469a909471424a285239d
If the encoder dynamic resize is triggered and change config()
is then called, it will reset the current (resized) codec width/height
back to the the config (unresized) width/height (which will then
prevent the resizing action from occurring in encoder_loop).
Avoid this by checking for a change in the config width/height
before resetting the cm->width/height.
Change-Id: Id9d50c0ee8a943abe4b6c72bbaa02d9696f93177
In VP9, the order for frame header was: [0] smooth, [1] regular, [2]
sharp, [3] bilinear. Per-block, the order was [0] regular, [1] smooth
and [2] sharp. For VP10, swap smooth/regular in the frame header so
that the block ordering and frame header ordering are interchangeable.
See issue #1046.
Change-Id: Ic9ec5964874375e40cd59bef50b489a76cbe4365
Unify the style of fdct4() fdct8() fdct16()
Add fdct32()
Add range_check() at each stage
Add unit test at ../../test/vp10_dct_test.cc
Change-Id: I9e912b2c5683862e65c5a21abc3e1c260cca4576
* changes:
test: limit the valid image size on OS/2
configure: add -Zhigh-mem to LDFLAGS on OS/2
configure: disable PIC on OS/2
Makefile: add $(STACKREALIGN) to CFLAGS for vp9_reconintra.c
x86inc.asm: fix NASM compilation
* changes:
Only build multithreaded functions on mt builds.
Don't build calc_psnr for high bit depth.
Enable missing dual lpf test
Remove unused VP10 functions.
Mark VP10 functions as 'INLINE'
Remove unused functions from test files
Only build append_negative_gtest_filter when it is used.
Add INLINE decoration to static test functions
* changes:
vp9_mcomp: make search functions private
vp9_mbgraph: use vp9_full_pixel_search(HEX)
vp9_temporal_filter: use vp9_full_pixel_search(HEX)
vp9_firstpass: make vp9_init_subsampling private
vp9_encoder: make vp9_alloc_compressor_data private
If either the encoder or the decoder is enabled, CONFIG_VP<N> will be
set. This simplifies the conditional and passes the chromium update
script when CONFIG_ values are passed in with 'yes' and 'no' values.
This was failing because it was checking against empty strings but
they are set to 'no'
Change-Id: I02ecd557210088ba1458cd0e89eead5666f6597a
Spatial/temporal svc code was removed. Verified using Borg test,
and the results before and after the change are matching.
Change-Id: I4c2ee5cd560428e3e50be02e57e5871ef4246390
For one pass CBR: only check for updating refresh_golden
if ext_refresh_frame_flags_pending is not set (i.e., == 0).
And move the resetting of ext_refresh_frame_flags_pending = 0
down to after the encode_loop (and account for dropped frames).
This is to prevent changing refresh_golden flga when the user
supplies the reference/update flags.
Change-Id: I4d87b3e705ba43f243667e367503b585c61e2a54
* changes:
Only build ssse3 filter functions on 64 bit
Clean up unused function warnings in vp8 encoder
Clean up unused function warnings in vp8 onyx_if.c
These were lost in the great sub pixel variance move of
6a82f0d7fb
Not having these functions caused a ~10% performance regression in
some realtime vp8 encodes.
Change-Id: I50658483d9198391806b27899f2c0d309233c4b5
previously any flags added while setting up the toolchain would
override the user selections; environment variables could be treated
similarly
Change-Id: Ibfcc644137d8e579af554d19a38d4020019a7a34
derflr +0.202%
hevclf +0.207%
hevcmr +0.095%
hevchr +0.077%
Tested locally on several derf sequences, speed (encoder + decoder)
is slower by less than 1%.
It is part of the EXT_TX experiment, which is to be continued to
explore different transform variants.
Change-Id: I05d44994a62106538a9a241ed8d89bd7c5d14761
Mark functions in findnearmv.h, invtrans.h and setupintrarecon.h
with INLINE.
Hide function in postproc.h behind the same #if as it's callers.
Change-Id: Ic1e014a943d2aca280f137019218b9d4f1443d61
This is a leftover of the XMA code which was removed a long time ago.
Found while looking for unused functions.
Change-Id: I07a3d542ae55440af59380dcdcf9a6c11cdfcb75
This commit adds clamp of new vectors similar to the logic in RD loop.
Such clamp is not necessary from the perspective of VP8 bitstream, but
is added to improve ChromeCast mirroring's robustness.
Change-Id: I42f6adbc60ffce283b994869364230858632d6fa
For each block in pickinter: use average of four middle
pixels (instead of single pixel) to set skin map.
This can help a little in reducing false skin detection in
some cases.
Change-Id: Ic247af75e9c2948b08ab977a39e061adacd8ec97
In high bitdepth setting, the rate multipier may be set as 0. In
lossless mode, the RD cost would always be 0, resulting in bad
partition and prediction mode choices.
Change-Id: I297014dd8bfa8a07ff0ab480119f75678300ff68
This patch just fixes the test for the time being, but does not
actually solve the underlying issue, which still needs investigation.
Change-Id: I54a35de839723f5b499b57e38dd2bdd400adc427
Switch to use the normative (convolve8) filter for source scaling,
only for 1/2x1/2 scaling for now. This is faster and has better
quality than either the vpx_scale_frame or the nonnormative scaler.
Remove the vp9_scale_if_required_fast, which is now not used.
Change-Id: I2f7d73950589d19baafb1fa650eac987d531bcc8
Define it as a function of reference frame types to provide
scalability for multiple reference frames.
Change-Id: I77b856c96916f352bc31004b9266b3f24e19bd0f
this restores the previous version's behavior avoiding issues with
builds that may split sources on directory boundaries; protected
visibility may work in this case.
Change-Id: If37c70d9bd81de85a8e112457b9819a5cac6129d
For 1 pass CBR mode under screen content mode:
if pre-analysis (source temporal-sad) indicates significant
change in content, then check the projected frame size after
encode_frame(), and if size is above threshold, force re-encode
of that frame at max QP.
Change-Id: I91e66d9f3167aff2ffcc6f16f47f19f1c21dc688
Only test for using golden as reference for variance partition
selection if it is used as a reference for that frame.
For temporal layers, golden may not be a reference on a given frame,
even though it was for some previous frame. If it is not a reference
for current frame, don't check/use it for partition selection.
Change-Id: I6b0f2bd36aebbb5903077c9a0a66d80f1de9a7b1
CONFIG_VP9_HIGHBITDEPTH is currently used by both vp9 and vp10, but in
many place outside vp9/vp10, the macro was used in conjunction of
CONFIG_VP9. This created a dependency on vp9 for vp10 to build. This
commit removes the dependency by use CONFIG_VP9_HIGHBITDEPTH only in
these places.
Change-Id: I8cc007fc9cf132394c6498ce6759e606b64a6ad0
This commit renames the vp10 encoder, decoder, and common interface
file names from vp9_ prefix to vp10_ prefix.
Change-Id: Iafb5d786e4b428d2b9bf097123bd86c4fa9ded24
For speed 7, real-time mode: Base layer frames are further apart
(for #temporal layers = 3, this is every 4 frames) so worth keeping
same motion search parameters (as in speed 6) on the base layer frames.
Change-Id: Idebf49dda6ef4f3d9a55aee55129a68253f692fb
gcc-based builds will allow a 0-element array, but visual studio builds
will not; this change hides the encoder and decoder specific symbols as
modules using them are selected based on the configuration.
Change-Id: Ic16ba9d12241070ec689dc5880164c14a4f7ca44
* changes:
Only use .text sections for aout
Use newer x86inc.asm
Use .text instead of .rodata on macho
Copy PIC handling code from x86_abi_support
Set 'private_extern' visibility for macho targets
Avoid 'amdnop' when building with nasm
Catch all elf formats
Expand PIC default to macho64 and respect CONFIG_PIC from libvpx
Use libvpx defines to set name mangling rules
Customize x86inc.asm for libvpx
Rename updated version of x86inc.asm
Use "private_prefix" instead of "program_name" and make vpx the default
prefix.
Change-Id: I4883a99b2aee8e5dc9f2c16a2e6f4b5d6e4de458
The read only sections are getting stripped on some OS X builds. As a
result, random data is used in place of the intended tables.
Change-Id: I58c18a53e503f093ee268451698c5761e6c32540
Other implementations of x86inc.asm have more comprehensive nasm
workarounds. This is the only thing that was changed for the previous
import to libvpx. See if we can still get away with it.
Change-Id: I3ef6fe9a4816461c89431a82b7e4a08b4b948d39
Make sure all variants get correct visibility and SECTION notes.
libvpx only pass elf32 and elf64 to the assembler, never just elf.
Change-Id: I7c36c115bf52436c9afe61985c859a2081948271
Revision a95584945dd9ce3acc66c6cd8f6796bc4404d40d
from git://git.videolan.org/x264.git
Temporarily name file x86inc.asm until all necessary local patches are
applied.
Change-Id: I9c7d0ed4d3ed900ae2d5db0abbcc048a2892c9b8
Use the correct period (in terms of cr->percent_refresh) for the condition
of larger delta-qp following key frame.
And account for larger interval for temporal layers.
Change-Id: Ibb43f5200f9b1eeb8bbb8211327b08ecda3c3b8a
Re-investigated the second-level sub-pixel motion search. Improved the
way of choosing search points. Rewrote the second-level search code.
At speed 0, the borg tests showed:
1. for stdhd set, Avg PSNR gain: 0.216%; Overall PSNR gain: 0.196%;
SSIM gain: 0.206%. Only 1 out of 15 clips showed PSNR loss.
2. for derf set, Avg PSNR gain: 0.171%; Overall PSNR gain: 0.192%;
SSIM gain: 0.207%. Only 3 out of 30 clips showed PSNR losses.
Added the condition for third-point checking, namely, less points
were checked. Speed tests showed no speed loss(Avg 0.3% speedup at
speed 0).
Change-Id: I6284ebb3fa7ba63be8528184c49e06757211a7f1
when configuring with mips32-android-gcc HAVE_MIPS32 would be set, but the
ndk does not set -mips32r2 for APP_ABI=mips which results in BSwap32 failing
to build; refine the check in endian_inl.h
Change-Id: I22893fe61f29111eb902d961b500b2174596268d
The test file compiler fails if one uses --disable-vp8-decoder
--enable-vp9-decoder. It effectively turns on CONFIG_VP8 and
CONFIG_DECODERS, but turns off CONFIG_VP8_DECODER, which causes
compiler error at test_vector_test.cc.
This commit fixes this issue by adding vp8/9 decoder flags to
the decoder behavior test, respectively.
Change-Id: I097ff8fd5e12715a94a565a82e54503885eb7187
-For ambient qp in active_worst setting: increase the initial
averaging time (from very first frame) to account for avg_qp of key_frame.
-In postencode on key frame: update the last_q/avg_q[key_frame] for
all temporal layers.
Change-Id: I5313153d350b1045b4835ce948dfffb7d2039b52
Condition usage of rc.frames_since_golden to non-svc mode.
rc.frames_since_golden, which is used in non-svc mode to add second reference,
was causing, under certain condiiton, the turning off of golden reference
for svc case.
Change-Id: Icec644d235d0471e56d8ff73d6c37278bd6ecd3b
and FUN_CONV_2D macros. The predict lut now handles
this case. The encoder now calls vpx_scaled_2d() instead
of vpx_convolve8() for scaling.
Change-Id: Ia1c8af8a31e4cb4887a587143108cb45835f7df7
This reverts commit a5e97d874b.
Additionally:
Revert "vpx_convolve_copy_sse2: fix win64"
This reverts commit 22a8474fe7.
This change performs poorly on various x86_64 devices affecting
performance by 1-3% at 1080P. Performance on chromebook like devices was
mixed neutral to slightly negative, so there should be minimal change
there.
Change-Id: I95831233b4b84ee96369baa192a2d4cc7639658c
This commit clears all the vp9_ prefix use case in vpx_dsp. It gets
the vp9 folder ready to branch out vp10.
Change-Id: I2906eec179ee792b4af8c9b4161313653050e931
This commit clears the function naming convention in vpx_dsp. It
replaces vp9_ prefix of global functions with vpx_ prefix. It also
removes the vp9_ prefix from static functions.
Change-Id: I6394359a63b71a51dda01342eec6a3cc08dfeedf
Choose a different diagonal point to check when the two costs are
the same, making it consistent with the way we choose the best mv.
This slightly changes the encoding result, and the derflr set borg
test at speed 0 shows 0.027% Overall PSNR gain, 0.024% Avg PSNR
gain, and 0.043% SSIM gain.
Change-Id: Ic8ee3a6767394866d159e4f9e1c777604dd73c17
If the current best mv(namely, the search center) is still the best mv
after the first level search, the second level checks is skipped. This
patch doesn't change the bitstream. At speed 0, it speeds up the encoder
by 1% - 2%.
Change-Id: I054c91b884d3f7aef157436c061744562bd6506d
Add a guard to exclud dspr2 inverse transform files from vpx_dsp
make file, when high bit-depth is turned on. This fixes the jenkins
nightly build.
Change-Id: Ibacd86563af1ec4810c550905b3fa0397baeeafc
Changes:
b6de61a Adds support for simple tags
75a6d2d sample_muxer: Don't write huge files.
cec1f85 mkvmuxer: remove unused timecode_scale variable
8a61b40 Merge "mkvparser: Tiny whitespace fix."
7affc5c clang-format re-run
d6d04ac mkvmuxer: use generic Cluster::AddFrame
4928b0b Merge "mkvmuxer: Write Block key frames correctly."
c2e4a46 Merge "sample_muxer: Use AddGenericFrame to add frames."
e97f296 mkvparser: Tiny whitespace fix.
d66ba44 Merge "Add support to parse DisplayUnit."
deb41c2 Add support to parse DisplayUnit.
42e5660 Fix issues on EBML lacing block parsing
fe1e9bb Fix block parsing to not allow frame_size = 0
2cb6a28 Change assertions to checks when parsing TrackPositions
d04580f Fixes issues on Block Group parsing
c3550fd mkvmuxer: Write Block key frames correctly.
5dd0e40 Merge "mkvmuxer: Set is_key to true for metadata blocks."
8e96863 mkvmuxer: Set is_key to true for metadata blocks.
a9e4819 sample_muxer: Use AddGenericFrame to add frames.
5a3be73 Change assertions to checks when load CuePoints
f99f3b2 mkvmuxerutil::EbmlDateElementSize: remove value param
ff572b5 Frame::IsValid: fix track_number check
b6311dc mkvmuxer: Refactor to remove a lot of duplicate code
256cd02 Merge "mkvmuxer: DiscardPadding should be signed integer."
16c8e78 mkvmuxer: s/frame/data in all AddFrame* functions.
c5e511c mkvmuxer: DiscardPadding should be signed integer.
4baaa2c Add framework build script: iosbuild.sh
3d06eb1 PATENTS: fix a typo: constitutes -> constitute
d3849c2 mkvparser: Dead code removal.
f439e52 Change assertions to checks when preloading Cues
d3a44cd Fix track transversal when listing Cues on sample
c6255af Tweak .gitignore so git status is clean after checkout and
build: - added missing underscore to sample_muxer - added cmake and make
related files
b5229c7 Makefile.unix: s/samplemuxer/sample_muxer/
e3616a6 Add support to parse stereo mode, display width and display
height in mkvparser
a4b68f8 parser: Fix bug in Chapters::Atom::Parse()
bab0a00 cmake: Set library and project name the proper way on Windows.
feeb9b1 Set library name to match Windows expectations.
b9a549b Fix CMakefile to generate libwebm.a
b386aa5 Add CMakeLists.txt and msvc_runtime.cmake.
b0f8a81 parser: Fix memory leak in Chapter parsing
f06e152 mkvmuxer: Fix MoveCuesBeforeClustersHelper recursive call.
27bb747 allow subtitle tracks with ContentEncodings
623d182 DoLoadCluster: tolerate empty clusters
1156da8 Update PATENTS to reflect s/VP8/WebM/g
0d4cb40 mkvmuxerutil: Use rand() in MSVC builds.
e12fff0 mkvmuxer: Overload WriteEbmlHeader for backward compatibility
a321704 mkvmuxer: write correct DocTypeVersion
574045e mkvmuxer: fix DiscardPadding
8be6397 Include crop elements when calculating size of Video element
8f2d1b3 mkvparser: fix DiscardPadding extraction
1c36c24 mkvmuxer: fix style guide violations
568504e Merge "UUIDs can have their high bit set"
acf788b Add support for CropLeft, CropRight, CropTop and CropBottom
elements.
418188b Merge "muxer: codec_id is a mandatory element"
07688c9 mkvmuxer: Reject frames if invalid track number is passed.
2a63e47 muxer: codec_id is a mandatory element
d13c017 UUIDs can have their high bit set
Change-Id: Iba28acb1ff774349d03e565f2641ddea132cf1e7
fixes link under vs9; this is the same change as:
dbf6e3f gen_msvs_vcxproj.sh: Avoid object name collisions.
Change-Id: I2a188c9024d0605e60e5e03ddcef1a25e7e53585
Chromium puts all the yasm output in the same directory. Looking at ways
to improve this but in the meantime get rid of collisions.
Change-Id: I923c5231d14e895ab96521eb89807ede868a0753
Ssim_vars is used to accumulate stats based 4x4 pixel blocks, this
commit changes the allocations size to be based on mi_rows and mi_cols
to avoid out-of-bound memory access for larger size videos. The hard
coded 720x480 can only work for image size up to 2880x1920.
Change-Id: Id9d07f3f777385b448ac88a6034b7472e4cf3c79
This commit moves the module inverse transform functions from vp9
to vpx_dsp folder. The hybrid transform wrapper functions stay in
the vp9 folder, since it involves codec-specific data structures.
Change-Id: Ib066367c953d3d024c73ba65157bbd70a95c9ef8
This got erroneously changed during the refactor. This fixes
SvcTest.TwoPassEncode2TemporalLayersWithMultipleFrameContextsAndTiles.
Change-Id: Ifa5ab0e098396c5e2d10478db87df256eadfa4c7
This function suffers from a couple problems in small core(tablets):
-The load of the next iteration is blocked by the store of previous iteration
-4k aliasing (between future store and older loads)
-current small core machine are in-order machine and because of it the store will spin the rehabQ until the load is finished
fixed by:
- prefetching 2 lines ahead
- unroll copy of 2 rows of block
- pre-load all xmm regiters before the loop, final stores after the loop
The function is optimized by:
copy_convolve_sse2 64x64 - 16%
copy_convolve_sse2 32x32 - 52%
copy_convolve_sse2 16x16 - 6%
copy_convolve_sse2 8x8 - 2.5%
copy_convolve_sse2 4x4 - 2.7%
credit goes to Tom Craver(tom.r.craver@intel.com) and Ilya Albrekht(ilya.albrekht@intel.com)
Change-Id: I63d3428799c50b2bf7b5677c8268bacb9fc29671
It in essence refactors the code for both the interpolation
filtering and the convolution. This change includes the moving
of all the files as well as the changing of the code from vp9_
prefix to vpx_ prefix accordingly, for underneath architectures:
(1) x86;
(2) arm/neon; and
(3) mips/msa.
The work on mips/drsp2 will be done in a separate change list.
Change-Id: Ic3ce7fb7f81210db7628b373c73553db68793c46
Don't run rate_block (cost_coeffs) if distortion alone is enough to
surpass best_rd.
This decreases 2nd pass runtime on HD at speed 2 by about 2%. There is
zero effect on output if tx_cache is removed.
Change-Id: Ia3b1cc77bfbe6ee988c395fde06c0eb92940b784
1. The RD scores obtained during the tx size selection were stored in the
tx cache, and used to help make the tx decision for the following frames.
This wasn't used anymore in VP9 encoder. Recovered the related decision
making code from 1.5+ years ago, and borg tests didn't show any quality
gain. This patch removed it to lower the complexity.
2. An optimization was done after the above refactoring. If the tx_mode
is not TX_MODE_SELECT, we only need to test the chosen tx size instead
of all posible tx sizes. This gave a 1.5% average speed gain at speed 2,
and a 1% average speed gain at speed 3.
Change-Id: Id8cd650e066a8cef33829d8c15388a8138adc78c
The forward 32x32 2D-DCT functions are aligned in vpx_dsp folder.
The vp9_dct.h file is not effectively used now.
Change-Id: Ie7946b6fdd784b8e91496242337bc9002c75c281
Replace the duplicate coefficient definition in neon implementations
of inverse transform with those from vpx_dsp/txfm_common.h
Change-Id: I4cd9bd9569ab1793dfdbb6f16d80bcb581599f0d
This commit replaces vp9_idct.h with txfm_common.h in many SIMD
implementation files for precise file dependency.
Change-Id: If73dd726bb16537e7494f28538b0a169810f9756
Separate the common coefficient constant into vpx_dsp/txfm_common.h.
Move the SSE2 macro definitions to vpx_dsp/x86/txfm_common_sse2.h.
This clears the use case of vp9_idct.h in vpx_dsp folder.
Change-Id: I319735a2abf42888e5080ac14cfbcde34be7b121
requires r10e or newer:
Android NDK, Revision 10e (May 2015)
...
Other bug fixes:
...
- Fixed .asm support for ABI x86_64.
Change-Id: I51ec9a5f77c982b7412d922e896348a83ae2d7d6
Avoid scaling the references if they have already been scaled.
Change only affects 1 pass non-svc mode for now.
Change-Id: I204f4079c026cba7adce7a7f855d072f6139ccec
The RD and load save/code grabs it as groups of four. In practice there
is no change to physical allocations becaquse this is backed by a 16-byte
memalign.
Change-Id: I01e89769872300e23227e03dd24a6e229f482025
Add vpx_dsp_rtcd.h to the header file list. The od_bin_fdct8x8()
here depends on forward 8x8 2D-DCT.
Change-Id: I1d71edc71f07069808823d2445c1cafd285e1b94
This commit factors out common macro definitions from the forward
and inverse transform implementations into vpx_dsp. It removes
the duplicate macro definitions from encoder and decoder folders.
Change-Id: I92301acbd3317075e9c5f03328a25abb123bca78
This commit factors the 4x4, 8x8, and 16x16 2D-DCT forward
transform operations into vpx_dsp folder.
Change-Id: I084b117b79c0925edcbcabb93f62b9f4bf8dbe7d
vp9_itrans*_dspr2.c aren't necessary for high bitdepth builds and
notably vp9_itrans8_dspr2.c fails in various configurations using a
codesourcery toolchain:
vp9_itrans8_dspr2.c:31:5: can't find a register in class 'GR_REGS' while reloading 'asm'
Change-Id: I2ac76203e65cc643cb835ab50e95701896d92a1a
This test places 128 in positions that would not be found
in the VP9 filter tables. The ssse3 code packs this table
into chars and uses the pmaddubsw instruction, which treats
the value as signed. The ssse3 code checks for 128 in
position 3, skipping the ssse3 code if found, and calls
vp9_convolve8_c(). vp9_convolve8_c() is also used for scaling.
ChangeFilterWorks breaks the ssse3 scaling code found in other
commits.
Change-Id: I1f5a76834bc35180b9094c48f9421bdb19d3d1cb
Eliminates the byte by byte read from bool decoder, by reading
in a size_t and then shifting it into place.
Change-Id: I0ed8c7b6f942847e79cc90105dc1d2b5b3deb0d6
This commit limits the scope of 1-D DCT and ADST functions within
vp9_dct.c and makes them static. This largely clears out the cross
referencing issue between vp9_dct.c and the SIMD optimizations.
Change-Id: If7cac478b11bb32328ccf70a9f60b709dad43d7f
The SSE2 version high bit-depth forward hybrid transforms are
essentially using the C functions via cross referencing to 1-D
functions in vp9_dct.c. This commit unifies the two versions and
removes the unnecessary dependency.
Change-Id: Ib4d0702a138f8daf7d0bd97c141ee7088f293765
Separate the hybrid transform case from 2D-DCT case. This will
allow us to clear up cross dependency between c and SIMD
implementations later.
Change-Id: Iaa499e8b096850a1c5a0c50a3b6e63e15d0184bf
The following quantization functions were moved:
vp9_quantize_b
vp9_quantize_b_32x32
vp9_highbd_quantize_b
vp9_highbd_quantize_b_32x32
vp9_quantize_dc
vp9_quantize_dc_32x32
vp9_highbd_quantize_dc
vp9_highbd_quantize_dc_32x32
The purpose of doing that was to allow these functions to be shared
by multiple codecs.
Change-Id: Id8ab939f283353cdd07bd930d47db3d932a5d87f
This commit moves the loop filter dspr2 implementation from vp9 to
vpx_dsp directory. It also fixes header file format issues.
Change-Id: I09203ed4bd267d7fd76bb79a6ee84a37646206b2
Remove the use of drop_frames_water_mark, as this is used for
frame dropping control. Use fixed threshold for now on buffer underflow.
Change-Id: If0ddda9f7f6fa96067cdcb0eccb42e17bda37c32
For vp9 decoder build without profile 2 and profile 3 support, this
commit changes to report error "Unsupported bitstream profile" for
input streams in profile 2 or 3, rather than other misleading error
information.
In addition, one of the invalid files in unit tests is actually coded
profile 2, this commit makes it tested only when the decoder is built
with vp9-highbitdepth.
This fixes issue #1028.
Change-Id: I8b6c1210787c8f89c703a546687dcf973ac20fc0
The various tap loop filter operations are common functions across
codec. This commit moves them along with SIMD optimizations to
vpx_dsp folder.
Change-Id: Ia5fa0b2e5289cdb98467502a549c380b9c60e92c
In aq-mode=3 under a resizing action (i.e., resize_pending != 0),
force an update of the golden reference frame.
Change-Id: I14806f6db71b5f8c827678cc5e1fc913c138a9a4
Fix bug in setting this flag for animated content.
The bug did cause quality to increase because far
more frames are not boosted than boosted.
However, the speed trade off to gain is a lot less
favorable and the behavior was not as intended.
Change-Id: I89fb70419c88b26f40b3534de0481730a1b3fcfa
Move the clamp functions to vpx_dsp_common.h file. Clear out the
dependency of vp9_loopfilter_filters.c on vp9_common.h file.
Change-Id: I9c4b928bcd7f597106b5aa96354356d3775a3431
Use drop_frames_water_mark for threshold on buffer underflow,
and change threshold for resize down.
Change-Id: I2de19adce50abe9bcdc0b107528cec8cc1857fcc
The fast scaling for 1 pass mode was being used only on the
first frame after resizing event (because resize_scale_num/den
is set to 1 and only changed for first frame following resize event).
Change-Id: I723b63e21823eb858f25f5662d2bbe4f1842e61f
Proper use/update of resize_state and resize_pending to constrain
the total amount of downsizing to be at most one scale down, for now.
Change-Id: Id18fc32499f2fbdbec16728dcdc9e4eac09098f0
From Change Ibf0c30b72074b3f71918ab278ccccc02a95a70a0
There is still an issue relating to one animated test clip with repeat
patterns where this change effectively increase the default maximum
arf interval by +1. This can be examined seperately.
Change-Id: Idd01d5480fc45202d8a059a0c3afc0997cc5bdd1
Flaten the intra block decoding process. It removes the legacy
foreach_transformed_block use in the decoder. This saves cycles
spent on retrieving the transform block position.
Change-Id: I21969afa50bb0a8ca292ef72f3569f33f663ef00
This commit simplifies the intra block boundary condition logic.
It removes the block index from the argument set.
Change-Id: If00142512eb88992613d6609356dfd73ba390138
Eliminates the byte by byte read from bool decoder, by reading
in a size_t and then shifting it into place.
Change-Id: Id89241977103fc3b973e4ed172a5cbf246998e5d
The clamp calls with INT32_MIN and INT32_MAX have no effect at all on
int values passed in, therefore this commit removes those effectless
clamps and also adds more const intermediate results to make the code
more readable.
Change-Id: I66d8811f58bb74ec31cbec9a6c441983a662352e
Rework the inter mode transform block decoding loop. Replace the
block index with the row and col index as the input argument. It
saves function call to compute the row and col index according to
the block index and overall block size, and many if statements
associated with the transform block position relative to the coding
block. For the test bit-stream pedestrian_area 1080p at 5 Mbps,
the decoding speed goes up from 81.13 fps to 81.92 fps.
Note that the intra coded block decoding needs more refactoring
work than the inter ones. So keep it using foreach_transforme_block
as for now.
Change-Id: I5622bdae7be28ed5af96693274057f55ba9b4fb4
The encoder gets its dqcoeff from the context tree. In the decoder move
it to directly after MACROBLOCKD.
Change-Id: I46c9b76f26956a360d17de0b26ecb994dae34ecb
Even if the recode loop is not enabled for the current frame type
trap the case where the projected size of a a frame is above the
maximum allowed in recode_loop_test()
Change-Id: I453004694b8f8699e3c2a83252e9f83adccdda4e
Changes to allow more use of rectangular partitions at
speeds 1 and 2 for content classed by the first pass as
animation and for blocks near the active image edge.
This has quite a big impact in quality for the animated
test sequence but also hurts encode speed for speed 2.
For other content types the impact on both speed and
quality is small.
Added some plumbing for detection of internal vertical
image edges.
Change-Id: I3fc48de2349f8cb87946caaf0b06dbb0ea261a9a
Change speed features / behavior for split mode when there
is an internal active edge (e.g. formatting bars).
Remove some threshold constraints in rd code near the active
edge of the image.
Add some plumbing for left and right active edge detection.
Patch set 5. Limit rd pass through for sub 8x8 to internal active edges.
This takes away any speed penalty for most clips but keeps the enhanced
edge coding for the more critical case of internal image edges
Change-Id: If644e4762874de4fe9cbb0a66211953fa74c13a5
Replace block index with transform type in the argument list. This
allows to save an extra fetch to the prediction mode. For pedestrian
area 1080p coded at 5 Mbps with single tile, the average decoding
speed goes up from 80.55 fps (before the refactoring series) to
81.13 fps.
Change-Id: Icbebf84ce63c19c0c92f3690ed201f6c3eab7881
If the pre-selected partition size (from variance partition) is
32x32, also apply nonrd partition search for 32x32 and 16x16 size.
Overall small positive gain in metrics, average ~1%.
Some visual improvement, for lower resolutions.
Change-Id: I69cb425bda94f7d13d34c451ab30e9276335a30e
The decoding process handles detokenization and reconstruction per
transform block sequentially. There is no need to offset the dqcoeff
buffer according to the transform block index. This allows to
reduce the memory spill and improve cache performance.
Change-Id: Ibb8bfe532a7a08fcabaf6d42cbec1e986901d32d
This commit replaces the vp8_ prefixed subtract function with the
common vpx_subtract_block function. It removes redundant SIMD
optimization codes and unit tests.
Change-Id: I42e086c32c93c6125e452dcaa6ed04337fe028d9
inline the code directly in read_mv_component(), the only place where it
was being used; this removes a function call in a hot function
Change-Id: I66f99c0c9ce3bc310101dbca4a470f023cc6fb55
To aid version management for integration with ffmpeg by use
of:
#ifdef VPX_CTRL_<CTRL_ID>
...
#endif
Change-Id: If550e06de4d3aa3685881f312ce6a86fa9de083b
Adds two new vp9 parameters --min-gf-interval and --max-gf-interval
to enable testing based on frequency of alt-ref frames.
Also adds a unit-test to test enforcement of min-gf-interval.
For both these parameters the default value is 0, which indicates
they are picked by the encoder, based on resolution and framerate
considerations. If they are greater than zero, the specified
parameter is honored.
(Additional note by paulwilkins)
Note that there is a slight oddity in that key frames are also GFs and
considered part of GF only group. However they are treated as not
being part of an arf group because for arf groups the previous GF is
assumed to be the terminal or overlay frame for the previous group.
(end note)
Change-Id: Ibf0c30b72074b3f71918ab278ccccc02a95a70a0
This reverts commit a42df86c03.
this change causes MSA/VP9SubpelVarianceTest.Ref and
MSA/VP9SubpelVarianceTest.ExtremeRef failures under
mips32r5el-msa-linux-gnu and mips64r6el-msa-linux-gnu
Change-Id: I40b71a0b774eaeb31f66f795733f95cf360909f7
This reverts commit 61774ad1c4.
this change causes MSA/VP9SubpelAvgVarianceTest.Ref failures under
mips32r5el-msa-linux-gnu and mips64r6el-msa-linux-gnu
Change-Id: I7fb520c12b2a3b212d5e84b7619a380a48e49bb0
The vp9_lpf_vertical_16_dual function optimized for x86 32bit target. The hot code in that function was caused by the call to the transpose8x16.
The gcc generated assembly created uneeded fills and spills to the stack. By interleaving 2 loads and unpack instructions, in addition to hoisting the consumer
instruction closer to the producer instructions, we eliminated most of the fills and spills and improve the function-level performance by 17%.
credit for writing the function as well as finding the root cause goes to Erik Niemeyer (erik.a.niemeyer@intel.com)
Change-Id: I6173cf53956d52918a047d1c53d9a673f952ec46
Added code to reduce the minimum partition size searched
for super blocks at or straddling the edge of the image.
If the first pass has detected formatting bars the "active" edge
may not be the real edge.
Change-Id: I9c4bdd1477e60f162a75fac95ba6be7c3521e05c
Correct the ARF boost calculations to partly discount
inactive or very low energy regions of the image.
Examples (formatting bars and 0 energy areas of animated clips).
Change-Id: I241af058d10aba8c67a4deca36deb913047d4561
This commit moves the primitive multi-threading files from vp9
folder to vpx_thread, which will be accessible by all vpx codec.
Change-Id: Ib51e66e9c69801c10631fab56d35a0c0aaed5883
the max value of the lookup in expanded form is:
(((1 << 7) - 1) << 1) - 65 + 1 + 64 = 254
remove the clamp [0, 253] and add one table entry
Change-Id: I0b5d0c66702fdb0b8f1cc9ab9b0dac66326e85a6
to MB_MODE_INFO_EXT. This saves 36 bytes per 8x8 area for
both the decoder and encoder. (encoder has two MODE_INFO
buffers)
Change-Id: If006abb2224acaf326df3c2be09e77e967662107
Only do the check for resizing if the feature is selected
(i.e., resize_mode = RESIZE_DYNAMIC).
And modify condition for checking to be resize_count >= window,
(since framerate can change).
Change-Id: Idceb4e50956bb965a1492b4993b0dcb393c9be4d
only uint8 is required; each use only loads one value as a uint8
quiets a few type conversion warnings
Change-Id: I03dc0dc0eb01ac23a6e8673daa2b77c6c57bf1b0
- Change default real time speed to -6.
- Add vpxenc_vp9_webm_rt_multithread, which encodes
niklas_1280_720_30.y4m with 2 to 4 threads using 2 to 4
tile columns.
Change-Id: I4d86c3360aec67ae5d1ba82eb6e0f0be8068b5af
Reduce boost for segment#2 for low bitrates and low-res.
This change is to reduce the rate overshoot at low bitrates.
No change in behavior, except at the very low bitrates.
Change-Id: I0dbd9d3b6356da5804de94adf10fca6a7a8f8948
Keep the same transform cutoff and partition selection
for speed 5 as in speeds >=6 (non-rd speed settings).
Existing setting for key frame at speed 5 allowed transform size
up to 32x32 on key frames, and did not allow for 4x4 block partition size.
This created more visual artifacts on first few frames.
avgPSNR/overallPSNR/SSIM gains of 0.2/0.7/0.8 for rtc_derf(low-res) set,
and 0/0.7/1.1 gains for rtc set.
Change-Id: I8c139ec6c9bb74e14b4ffbad5f12e94f18a59c0b
configure.sh was setting some Mac OS X options for iOS targets, which
confuses the iOS 9 beta SDK in Xcode 7 when linking libraries.
Additionally, old armv6 media extensions were being enabled on iOS
when they're not needed (we always have Neon since iOS 6). These
broke on iOS 9 SDK which no longer assembles those instructions.
Change-Id: I4e4d2722392ead3382ce96289c03ef1e489799d6
skips testdata verification; useful with slow media or if the data was
retrieved via a separate call to testdata
Change-Id: Ifd97892cee6c04b0111874cc8071675e90ec852b
For speed 5 real-time mode, the selection of the partition size for
superblocks on the segment (aq-mode=3) uses the non-rd recursive
pick partition search, and can sometimes select 64x64.
For low resolutions, visually better to limit this to 32x32.
Change-Id: I69657a7ed8899f8b3cf8c9c318a2509c5c72c565
For screen content don't refresh a block at a quantizer higher than
it was last coded at. PReviosuly at realtime speeds the encoder had a
tendency to recode a block from GOLDEN with a higher Q than it was last
coded at.
Change-Id: Iacd561806c769dcce1a81b9827ffc70090f5ba18
Decision to scale down/up is based on buffer state and average QP
over previous time window. Limit the total amount of down-scaling
to be at most one scale down for now.
Reset certain quantities after resize (buffer level, cyclic refresh,
rate correction factor).
Feature is enable via the setting rc_resize_allowed = 1.
Change-Id: I9b1a53024e1e1e953fb8a1e1f75d21d160280dc7
There is a naming conflict in the chromium build system.
The rest of the variance functions will move to vpx_dsp soon.
Change-Id: Iff78da2aafb0d7380eda73e38d7dac72110a1e47
The internal behavior of block_yrd differs in high bit depth
settings from 8-bit one. This causes the assertion condition not
true for high bit depth.
Change-Id: I15dc02e7162d27cabe78c451941d769d488b1174
The overflow issue affects a variable that is only used in inter
mode. This commit fixes the ioc warning triggered in the intra
mode. It does not affect the compression performance.
Change-Id: I593d1b5650599de07f3e68176dd1442c6cb7bdbc
set_frame_size() is being called twice, once before entering
encode_encode_frame_to_data_rate(), and once again in that function.
No need to call it twice for one-pass mode.
Change-Id: I5fabaf0a90482d4f42cd89ef7ae1402c31aec600
This patch modified the thread creating code. When use_svc is true,
the number of threads created is decided by the highest resolution.
This resolved WebM issue 1018.
Change-Id: I367227b14d1f8b08bbdad3635b232a3a37bbba26
* changes:
vp9_decodeframe.h: remove unused prototype
vp9_decodeframe: move public funcs to end of file
vp9_decodeframe: reorder some functions
vp9_decodeframe: hide vp9_dec_build_inter_predictors_sb
This commit fixes a potential integer overflow issue in function
hadamard_16x16. It adds corresponding dynamic range comment.
Change-Id: Iec22f3be345fb920ec79178e016378e2f65b20be
the declaration containing the alignment in vp9_filter.h was removed in:
eb88b17 Make vp9 subpixel match vp8
fixes a crash in 32-bit builds
Change-Id: I9a97e6b4e8e94698e43ff79d0d8bb85043b73c61
If the frame size increases, the tile data buffer needs to be
re-allocated according to the number of tiles existing in current
frame. This patch makes the multi-tile encoding work in spatial
SVC usage case, and partially solved WebM issue 1018.
Change-Id: I1ad6f33058cf5ce6f60ed5024455a709ca80c5ad
vp9_init_dequantizer() was deleted in:
bdd249b Optimize the dequantization process on decoder side.
Change-Id: Iedb5b6a3a03964dd6901c1e3b2325194d94bc708
add a dependency on *_rtcd.h to ensure they're generated before
attempting to build the test files
Change-Id: Ibbbd1f6ea77912bfd297129e7c83b9a80923ea12
Reduce motion threshold and boost factor for second segment,
for low bitrates, at low resolutions for now.
This is to reduce the rate fluctuation/frame dropping that occurs
at these low bitrates.
Change-Id: Ia66c3be41831882fca8c1e4fe104f5ea8fbe7142
Some initial experiments into discounting dead zone
formating bars and intra skip blocks (common in some
types of animation and graphics) in the calculation of
the active max Q for each ARF/GF group.
TODO: check for vertical formating bars and validate the
horizontal bar at the bottom edge of the image.
As expected, this change as it stands, does not make much
difference for the natural videos in the std-hd and derf sets.
However, for the yt and yt hd set there is a significant rise
in the average PSNR with overall PSNR and SSIM remaining
neutral.
The mean rise for the YT-HD test set was > 6%. This is mainly
because the change allows Q to drop further on titles and
other graphics sections where spending a small number of
extra bits gives a sharp rise in PSNR.
Change-Id: I3f878ae91fc1854312d7ecf9fa792c17bc1aa6b7
For content that is identified as likely to contain some
animation or graphics content, increase the availability
of split modes for good quality speeds 1-3.
On a problem test animation clip this improves metrics
results by about 0.25 db and makes a noticeable difference
visually. It also causes a small drop in file size (~0.5%) but
a rise in encode time of about 5-6% at speed 2.
For more normal content it should have no effect.
Change-Id: Ic4cd9a8de065af9f9402f4477a17442aebf0e439
Added check to see if last frame was all intra. This will
eliminate two checks in find_mv_refs_idx(). Also, do not
update the frame mvs if the current frame is all intra.
This improved performance on material with frequent
intra-only frames.
Change-Id: I44a4042c3670ab0d38439d565062a0e2a1ba9d1e
fails unit tests:
[ FAILED ] NEON/VP8SubpelVarianceTest.ExtremeRef/0, where GetParam() = (3, 3, 0x14e36d, 0)
[ FAILED ] NEON/VP8SubpelVarianceTest.Ref/0, where GetParam() = (3, 3, 0x14e36d, 0)
the tests were recently enabled in:
eb88b17 Make vp9 subpixel match vp8
the functions likely haven't changed since being converted from assembly
Change-Id: I6141717b111b8f735f436c160d74270af53ef722
move them under their respective config check to avoid some unused
variable warnings when disabled
Change-Id: Ic5e5280cf1bc1f56e8349676f0bedae4acef34ea
this quiets warnings from armv6 code [1].
from msdn [2]:
-oldit
Generate ARMv7-style IT blocks. By default, ARMv8-compatible IT blocks
are generated.
a new configuration would be needed for armv8 in any case as the neon
assembly is being built, so removing this should be harmless
[1] A4509: This form of conditional instruction is deprecated
[2] https://msdn.microsoft.com/en-us/library/hh873189.aspx
Change-Id: I4c3b838b52a87401c6daecd83d22ab148ed7c5d9
This control allows the application to skip the loop filter in the
decoder. This is an advanced control that should only be used in
extreme circumstances as it may introduce and accumulate decode
artifacts.
Change-Id: I278c65c60826f84c9141ebe06c6eeed3c2335fa8
WebM files will adjust the display width and height according to the
input pixel aspect ratio. The default pixel aspect ratio is 1:1.
BUG=https://code.google.com/p/webm/issues/detail?id=1005
Change-Id: I23e0a601b7259fa9513cb86110c41b8437769808
calculate the averages needed for even and odd rows once; this removes a
conditional from the inner loop
the final average calculated currently relies on above[] being extended,
it could be reduced to use
above[block_size - 2] + 3 * above[block_size - 1]
Change-Id: I70f5eac8d8a2a959c7114844a95826f445c3dd4d
The only difference between the two was that the vp9 function allowed
for every step in the bilinear filter (16 steps) while vp8 only allowed
for half of those. Since all the call sites in vp9 (<< 1) the input, it
only ever used the same steps as vp8.
This will allow moving the subpel variance to vpx_dsp with the rest of
the variance functions.
Change-Id: I6fa2509350a2dc610c46b3e15bde98a15a084b75
Adds code to detect dead zone bars at the top and bottom
of reformatted letterbox video (note that the code only
looks at the top of the image and assumes any dead zone
is symmetrical). Use of this to adapt rate control etc.
will follow in a subsequent patch.
Also counts other blocks (excluding the dead zone) that
have no intra signal. The presence of a significant
number of such blocks can be used as a identify that the frame
may be artificial (e.g. animation, screen capture, graphics).
This patch contains plumbing only and does not use
the signal.
Change-Id: I59bc93529cd4065416cef773e405fda3ae006a20
Some places are using the unoptimized variance function. This was never
intended and does not fit into the optimization framework.
Change-Id: Id96238407aad03b0ffd4a46cd183555a026daedc
Updated sources according to improved version of common MSA macros.
Enabled respective convolve MSA hooks and tests.
Overall, this is just upgrading the code with styling changes.
Change-Id: If5ad6ef8ea7ca47feed6d2fc9f34f0f0e8b6694d
Clang adds alignment hints when casting up the loads/stores. Although
this should be safe for most paths, it's causing some crashes. Either
the source of the misalignment needs to be determined and adjusted or
the intrinsics need to be rewritten to avoid using the cast to load the
data.
BUG=817,892
Change-Id: Ia3aa824d6a4cd97e14325ff49dc730b6f85ec7e8
The larger internal variables are required for the intermediates
but RoundHighBitDepth brings them down to uint32_t/unsigned int.
Fixes type warnings in visual studio.
Change-Id: I48d35284d6cbde330ccdc1f46b6215a645d5eb00
Numerator was being range checked against the
denominator - preventing any frame rate slower
than 1 fps.
I've tested this on a Mac using using ffmpeg and
results are comparable to mp4 and ogg files generated
at the same time.
Not yet tested on Windows.
Johnny Klonaris
google@jawknee.com
Change-Id: Idb358dbc2e7dc000037880ede4a1b0df248a42c8
Updated sources according to improved version of common MSA macros.
Enabled idct MSA hooks and tests.
Overall, this is just upgrading the code with styling changes.
Change-Id: I1f488ab2c741f6c622b7a855388a202168082209
only the immediate above right pixel is needed; this removes a
conditional from the inner loop
the final average calculated currently relies on above[] being extended,
it could be reduced to use above[block_size - 2] + 3 * above_right
Change-Id: Ica4f2b8d25eec3ca1d6fa52ef0d4adc228eeea3f
Done little restructuring/styling changes to the sources like generic macro definitions, their use to reduce code lines, better code alignments etc.
Disabled all MSA hooks and tests
Change-Id: Ic6f2dce0b501f46b80c06c46c0fe2043d557b190
Keep the logic, transform size based on cyclic refresh and bsize,
(that was conditioned on VAR_PARTITION conditions) the same
for all speeds in non-rd mode (speeds >= 5).
No change to speeds >=6.
Small improvement for speed 5, ~0.5/1.5% gain for avg psnr/ssim.
Change-Id: If9c5657f3d30efd3c7f147166bba7cb69ea55114
In VS 2015 and higher snprintf is supplied and therefore vsnprintf
doesn't need to be defined. This also avoids problems caused by
_snprintf being different from snprintf.
This fixes a build break with VS 2015 and improves security.
Originally submitted via chromium by brucedawson@chromium.orghttps://codereview.chromium.org/1055603003
Additionally break this MSVC-specific tweak to a new file, which will
become the home of all such MSVC-specific things.
This requires adding a dependency on msvc.h to every example which uses
args.c and tools_common.h
Change-Id: I35b5f8e7ea00f6627403aabc9ea79b0412557a99
ROUND_POWER_OF_TWO has some poor side effects when used
with [u]int64_t such as doing the shifting in 32bits.
Change-Id: Ic85a19765cd316fb43657cb21c86f35ceb772773
Increase the 32x32 split threshold, to allow for more 32x32
at expense of 16x16. Visually looks somewhat better.
Change-Id: Ia1439c3a0dc2d7933468b88bd59266fcd9f03505
Break out the setting of the block variance split thresholds,
since they are locally modified, e.g., based on local/segment qp.
No change in performance.
Change-Id: I0a3238e6dab05140657539fc4bd27ac5ff7a554e
+ synchronize filter function signatures
this makes any intrinsics filters available for inlining and has the
side-effect of making those filters static, quieting missing-prototype
warnings.
Change-Id: I1908875caffa585bd4fc65aaf10d17a5e20cfb46
+ synchronize filter function signatures
this makes any intrinsics filters available for inlining and has the
side-effect of making those filters static, quieting missing-prototype
warnings.
Change-Id: I1cd55c9d52547793ad65aa90c7620f0e426edaa2
collect the vp9_convolve function definition macros there; this will
allow some relocation of functions from vp9_asm_stubs.c
Change-Id: Idadd117fa256dd48748379856973fd985b8204e8
This commit fixes the integral projection motion search crash when
frame resize is used. It fixes issue 994.
Change-Id: Ieeb52619121d7444f7d6b3d0cf09415f990d1506
reorder includes to avoid:
warning C4985: 'ceil': attributes not present on previous declaration.
this is the same workaround used in vp9/common/vp9_systemdependent.h
Change-Id: Ia10dd63de24f96fa1507a6179220e9d6ec774db6
Various header/test files had to be re-worked in order to
build "Remove cm parameter from vp9_decode_block_tokens()".
This patch reverts the "Remove cm" part and only contains
the re-worked header files.
Change-Id: I520958a88d1991fee988a3c784d0eac40e117a32
1. Check existing buffer sizes when re-allocate context buffers.
2. Don't need to set mi buffers to 0 during setup_mi.
Change-Id: I6b48b0e077a4d804312b605ad0dc34aec5795a6d
This patch provides a partial rapid feedback of bits
resulting from extreme undershoot.
Some improvement on some problem animated material
but in its current form only a small impact on the metrics results
of our standard test sets.
Change-Id: Ie03036ea8123bc2553437cb8c8c9e7a9fc5dac5d
This patch addresses two issues that can occur when the
encoder chooses to use a mixture of ARF and GF groups.
The first issue relates to a failure to reset the "ARF active" flag
correctly when transitioning from coding ARF groups to coding
GF groups. This caused some golden frames to be encoded
with an incorrect bit rate target as if they were ARF overlay frames.
The second issue relates to the encoding of a single short GF group
just before a key frame. Where the last group before a key frame
is an ARF group we expect the final frame before the key frame to
be an low data rate overlay frame. However, when the last group
is a GF group, the final frame before the key frame should be a normal
frame with a normal bit allocation. This issue had the potential to cause
a single poorly coded frame just before a key frame. If that key frame
were a forced key frame rather than a real scene cut, this might cause
pulsing.
Change-Id: Idf1eb5eaf63a231495a74de7899236e1ead9fb00
This allows rate control to react to content of current frame being encoded.
Enabling this feature via the setting: screen_content_mode = 2.
Change-Id: Ib2c6670551d96f4907495d5b7b76bb8c49e673db
this allows test_libvpx's simd caps check to be used; it also fixes a
link error on OS X with -fcommon.
Change-Id: I1a62a3e74ba06b8f3b37a22fcfdebf90c04ab289
in addition to <arch>/*. this will pick up tests defined with TEST()
instead of INSTANTIATE_TEST_CASE_P()
Change-Id: I0917741baac89d9ce857f4d4aa53790e8a0c6c12
split call of extend_and_predict() and return, fixes visual studio build
warning since:
0a80164 Move mc_buf to cut down size of MACROBLOCKD.
Change-Id: I7cdf712941ef773a07f038539cb8080dc27861cd
this file shouldn't be built directly, it is included in vp9_dct_sse2.c
to create a non-high-bitdepth and a high-bitdepth version
silences missing prototype warnings for the unused FDCT* functions
Change-Id: Ide6ff8c24ab31bdb0f833260505ae33660a1ad5b
this file shouldn't be built directly, it is included in vp9_dct_sse2.c
to create a non-high-bitdepth and a high-bitdepth version
silences missing prototype warnings for the unused FDCT32x32* functions
Change-Id: I0e38f16dae5ea1728de184ee2c89287d48675c51
this file shouldn't be built directly, it is included in vp9_dct_avx2.c
to create a non-high-bitdepth and a high-bitdepth version
silences missing prototype warnings for the unused FDCT32x32* functions
Change-Id: I4c19935c0e035b393be513bde735e9a78064a494
* changes:
vp9_subexp.h: add a missing prototype
vp9: add some missing includes
vp9 intrinsics: add vp9_rtcd include
vp9: correct some function signatures
vp9_variance_sse2: sync function signatures
vp9/encoder: make some functions static
vp9_dct_sse2: make some functions static
vp9_decodeframe.c: make a function static
Use the same settting as in speed >=6.
This will use same logic for tx_size selecton as in speed >=6,
which limits the transform size and reduces ringing artifact.
Also metrics go up on average with this change: ~2% for PSNR, ~10% for SSIM.
Change-Id: Ia2d50db236ae1cc72f742bfa6c9ec5ea50ff0e0a
* changes:
vp8/rdopt.h+onyx_int.h: add some missing prototypes
vp8: add some missing includes
vp8: make some functions static
vp8/common/variance*: add vp8_rtcd include
vp8_copy32xn: sync function signature
useful for speed testing / verifying individual function optimizations;
currently tests non-high-bitdepth VP9 intra predictors
Change-Id: Ibd247765e43a31894697d43f1d39d312e0ba2090
Add a check to make sure we have a decoded frame available
before copying its 'corrupt' flag.
(Originally submitted to the old repository by Alexander Voronov
as: https://gerrit.chromium.org/gerrit/#/c/74305/).
Change-Id: Iceb4686c785afb437b668015bf8818b18d60e0ce
Testing on another rate control patch reveals that in some
situations, where the encoder is flipping in and out of arf
mode, we get an encoder decoder mismatch.
Whilst it is still not clear why, skipping the last buffer
update seems to trigger the problem. Until I can establish
why, or if there is another underlying cause, I am reverting
this change.
This reverts commit e5112b3ae3.
Change-Id: I315c5200414de89458015823344b7367e9dd75ba
With the sad functions, and hopefully the variance functions soon,
moving to the vpx_dsp location, place the defines used in the
reference C code in a common location.
Change-Id: I4c8ce7778eb38a0a3ee674d2f1c488eda01cfeca
Basically just a warning, but disconcerting nonetheless. Removes this
output from the build:
Makefile:59: -x86_64-darwin13-gcc.mk: No such file or directory
Change-Id: Ibb379506352b2f613ef4a7b1ac47e9c95d0d1580
These targets no longer build (PPC support was removed from
libvpx). Remove the dead code and misleading help output.
BUG=https://code.google.com/p/webm/issues/detail?id=997
Change-Id: Ib35614806adeae970f3821da0d8dbcc54ab8d868
The computation of new metrics is not supported yet in highbitdepth
mode. This commit adds protection to make sure the computation is
done only when highbitdepth is not on. This protection shall be
revised when support of highbitdpeth computation is added.
This resolves the encoder crash when configured with both
--enable-internal-stats
--enable-vp9-highbitdepth
Change-Id: Id9f4bcc4fa26d9ca0e9eabade83f3f88a5b212e6
This patch fixed the following warning:
src\third_party\libvpx\source\libvpx\vp9\encoder\vp9_pickmode.c(1607) :
warning C6246: Local declaration of 'this_mode' hides declaration of the
same name in outer scope.
Change-Id: I1d93c4a47a13cb13089fec5bd61e8b58e6cd8d58
* changes:
vpxenc: make some functions static
vpxdec: make some functions static
tools_common.h: fix get_vpx_decoder_count() proto
tools_common.h: fix get_vpx_encoder_count() proto
tools_common.h: fix usage_exit() prototype
rename LIBVPX_TEST_BINS to LIBVPX_TEST_BIN and remove foreach usage.
this was a leftover from having multiple targets with their own (single)
object list; the use of LIBVPX_TEST_OBJS so widely makes extending these
loops difficult.
Change-Id: I61bda1b91acb43145609f04b8fe6e45ec4483e22
When aq-mode=3 is enabled, only for base layer frames should the
qp of the frame incorporate the segment delta-qp.
This was causing more rate mismatch for the enhancement layer frames
when running temporal layers with aq-mode=3 on.
Change-Id: I1c5e69d1ef8a51188af8696753c17fd8f67699b3
currently this needs to be 2x (NEED_ABOVERIGHT) the size of the largest
block (32) + 1 (for above_left). reduce the buffer size from 128 + 16
(alignment) to 64 + 16.
Change-Id: Idaca1806c7e1214e9437de24e15edc2ebf18f95d
The warning only happens in VP9 encoder's first pass due to src_mi
is not set up yet. But it will not fail the encoder as left_mi and
above_mi are not used in the first_pass and they will be set up again
in the second pass.
Change-Id: I0713b4660d71e229e196654cb0970ba6b1574f28
Where a frame appears to be a repeat of an earlier
frame or frame buffer, but the first pass code
does not anticipate this (usually because it is matching
the GF or ARF buffer not the last frame buffer), do not
update the last frame buffer.
This helps ensure that the content of the last frame buffer
is kept "different" where possible, and not updated to
match the GF or ARF. This is particularly helpful in some
animated sequences where there are groups of repeating
frames. Here it has quite a big impact. However, in most
of our standard test clips it has little or no impact.
Change-Id: I77332ee1a69f9ffc0c6080bfeb811c43fd8828e6
this macro was used inconsistently and only differs in behavior from
DECLARE_ALIGNED when an alignment attribute is unavailable. this macro
is used with calls to assembly, while generic c-code doesn't rely on it,
so in a c-only build without an alignment attribute the code will
function as expected.
Change-Id: Ie9d06d4028c0de17c63b3a27e6c1b0491cc4ea79
Using EIGHTTAP and EIGHTTAP_SMOOTH seem sufficient.
Hard to see any visual gain from allowing EIGHTTAP_SHARP, and it is
rarely selected.
PSNR/SSIM metrics go up by ~0.18/0.14%.
Change-Id: I96fa0d98f9321b913e3ebcd464d4ff3c63018791
Create a new component, vpx_dsp, for code that can be shared
between codecs. Move the SAD code into the component.
This reduces the size of vpxenc/dec by 36k on x86_64 builds.
Change-Id: I73f837ddaecac6b350bf757af0cfe19c4ab9327a
Added the intra mode early termination in order to
speed up the mode search in non-rd case since we
started to include more intra modes in the search
list. Borg tests(rtc set) showed a 0.048% PSNR gain
and 0.061 SSIM gain. No speed change.
Change-Id: I6f255fe534dc50b736e6a66a726ad458eb9b4443
widen the loads and stores to 128-bit.
this was added, but not enabled in:
493a857 Add some sse2 code for intra prediction.
Change-Id: I277d7db608a7db7d75cc0bde86f48fa66ad487e4
For non-rd mode (speed >=5): use mask based on prediction block size, and
(for non-screen content mode) allow for checking horiz and vert intra modes
for blocks sizes < 16x16.
Avg psnr/ssim metrics go up by about ~0.2%.
Only allowing H/V intra on block sizes below 16x16 for now, to keep
encoding time increase very small, and also when allowing H/V on 16x16 blocks,
metrics went down on a few clips which need to be further examined.
Change-Id: I8ae0bc8cb2a964f9709612c76c5661acaab1381e
Impose a limit on the rd auto partition search based on
the image format. Smaller formats require that the search
includes includes a smaller minimum block size.
This change is intended to mitigate the visual impact of
ringing in some problem clips, for smaller image formats.
Change-Id: Ie039e5f599ee079bbef5d272f3e40e2e27d8f97b
Remove one of the auto partition size cases.
This case can behaves badly in some types of animated content
and was only used for the rd encode path. A subsequent patch
will add additional checks to help further improve visual quality.
Change-Id: I0ebd8da3d45ab8501afa45d7959ced8c2d60ee4e
Previously limit on max interval set to 0.5 seconds.
Though this helped some low frame rate material it
appears to be a bit too aggressive for some 24 and 25 fps
content. This patch relaxes the limit to 0.75 seconds.
The patch also adds a new minimum interval variable
to replace the current hard wired value. This allows us
to impose a limit on the maximum number of primary
arfs per second for high frame rate (e.g. 50 & 60fps)
content. This is to address concerns regarding playback
performance on some platforms if there is a high base
frame rate and very frequent arfs.
Change-Id: I373e8b6b2a8ef522eced6c6d2cceb234ff763fcf
offsetting by a variable stride prevents instruction reordering,
resulting in poor assembly.
additionally reroll 16x16/32x32 loops to reduce register spill with this
new format
Change-Id: I0635b8ba21ecdb88116e927dbdab53acdf256e11
This patch reduced the BLOCK_8X8's intra_cost_penalty, which
allows 8x8 blocks to conduct intra mode search. Borg test
result(rtc set): 0.077% PSNR gain, 0.228% SSIM gain. No speed
changes.
Change-Id: Icfe90c4f6969de24bda8ecacbd3da50330bf22b2
The rotation computation using 2X of cos(pi/16) has a potential to
overflow 32 bit, this commit disable the function to allow further
investigation and optimization.
Change-Id: I4a9803bc71303d459cb1ec5bbd7c4aaf8968e5cf
Calculated cpi->vbp_threshold_sad from this frame's dequant value.
The encoding quality and speed didn't change much. Borg test
result: PSNR: -0.002%, SSIM: -0.003%.
Change-Id: I97c9826986f39582f29910d637d08a69c90afdee
The version is currently producing different result from c version
for some input. Disable the use of it for now to allow time for
investigation the source of mismatch.
Change-Id: Id039455494ee531db4886a9f1fa4761174ef6df3
The default golden frame interval was doubled. After encoding a
frame, the background motion was measured. If the motion was high,
the current frame was set as the golden frame. Currently, the
changes were applied only while aq-mode 3 was on.
Borg tests(rtc set) showed a 0.226% PSNR gain and 0.312% SSIM gain.
No speed changes.
Change-Id: Id1e2793cc5be37e8a9bacec1380af6f36182f9b1
structured extended feature flags require eax = 7; avoids incorrectly
detecting avx2 on some older processors that support avx.
from [1]:
INPUT EAX = 0: Returns CPUID’s Highest Value for Basic Processor
Information and the Vendor Identification String
[1] http://www.intel.com/content/www/us/en/processors/processor-identification-cpuid-instruction-note.html
Change-Id: I6b4735b5f7b7729a815e428fca767d1e5a10bcab
For color sampling format other than 420, valid partion size in Y may
not work for UV plane. This commit adds validation of UV partition
size before select the partition choice.
This fixes a crash for real time encoding of 422 input.
Change-Id: I1fe3282accfd58625e8b5e6a4c8d2c84199751b6
(see I3a05cf1610679fed26e0b2eadd315a9ae91afdd6)
For the test clip used, the decoder performance improved by ~2%.
This is also an intermediate step towards adding back the
mode_info streams.
Change-Id: Idddc4a3f46e4180fbebddc156c4bbf177d5c2e0d
The existing test was triggering a lot of false positives on some types
of animated material with very plain backgrounds. These were triggering
code designed to catch key frames in letter box format clips.
This patch tightens up the criteria and imposes a minimum requirement
on the % blocks coded intra in the first pass and the ratio between the
% coded intra and the modified inter % after discounting neutral (flat)
blocks that are coded equally well either way.
On a particular problem animation clip this change eliminated a large
number of false positives including some cases where the old code
selected kf several times in a row. Marginal false negatives are less
damaging typically to compression and in the problem clip there are now
a couple of cases where "visual" scene cuts are ignored because of well
correlated content across the scene cut.
Replaced some magic numbers related to this with #defines and added
explanatory comments.
Change-Id: Ia3d304ac60eb7e4323e3817eaf83b4752cd63ecf
vestigial. the code is stale and couldn't be configured directly; there
are better ways to achieve this now
Change-Id: I5a9c62e099215588cd0d7e5ae002dfc77c21a895
PSNR HVS is a human visual system weighted version of SNR that's
gained some popularity from academia and apparently better matches
MOS testing.
This code is borrowed from the Daala Project but uses our FDCT code.
Change-Id: Idd10fbc93129f7f4734946f6009f87d0f44cd2d7
When the tokenization is not taking effect, the tokenization
pointer remains unchanged. No need to re-assign the backup pointer
value.
Change-Id: I58fe1f6285aa3b4a88ceb864c11d5de8ac6235dd
This partially reverts commit 14ef4aeafb
Including the rtcd headers to get the function definitions causes
problems on VS9.
Change-Id: I780874d9e03af2d3124192ab0e3907301f22674c
This patch limits the maximum arf interval length to
approximately half a second. In some low fps animations in
particular the existing code was selecting an overly long interval
which was hurting visual quality. For a sample problem test clip
(360P animation , 15fps, ~200Kbit/s) this change also improved
metrics by >0.5 db.
There may be some clips where this hurts metrics a little, but the
worst case impact visually is likely to be less than having an
interval that is much too long. On more normal material at 24
fps or higher, the impact is likely to be nil/minimal.
Change-Id: Id8b57413931a670c861213ea91d7cc596375a297
nothing is using android/log.h currently; also quiets a warning when
building a static lib:
Android NDK: WARNING:libvpx/build/make/Android.mk:vpx: LOCAL_LDLIBS is
always ignored for static libraries
Change-Id: I1469a5d6fca4f7338a4a82e26a03e60fc91d92ca
for big endian disable msa
removed -flax-vector-conversion flag
disable runtime_cpu_detect feature if enabled
Change-Id: Icd5130b733f2ddcdb94ffb6e4d170e6ca0f9832b
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
Refactor the loops in dec_build_inter_predictors to try and decrease
the number of instructions. Limited testing saw about 1% perf
increase on x86 and about 0.67 % perf increase on Arm.
Change-Id: I69cfe6335bb562fbaaebf43fb3f5c5a2a28882a2
add a check for the status line to awk and better report failure given
the program output will be lost in this case
Change-Id: I1348a80108c81099d609f2e2227dd2c31bd8cd54
Modifies a special handling that improves rate control accuracy in
the constrained quality mode, when the undershoot and overshoot
limits are set tighter.
Change-Id: If62103f0ef3ed1cac92807400678c93da50cf046
the sse4 code expects 16-byte aligned arrays; vp8 already had a similar
change applied:
b2aa401 Align SAD output array to be 16-byte aligned
Change-Id: I5e902035e5a87e23309e151113f3c0d4a8372226
Skip redundant prediction filter type cost in filter search loop,
if the rate value will be reset in Hadamard transform based rate
distortion estimate.
Change-Id: Ie5221f4bc8da9461c449df367251aeeac52c6e5d
Reset the reached_eos flag in webm_guess_framerate in case it ends
up consuming the entire file. Also adding a vpxdec shell test to
verify this behavior.
Change-Id: I371eebd2105231dc0f60e65da1f71b233ad14be5
* changes:
vp8_regular_quantize_b_sse2: remove dead init
vp8cx_pick_filter_level*: remove dead inits
vp8_decode_frame: remove dead increment
rdopt: remove dead stores
find_next_key_frame: remove dead init & store
multiframe_quality_enhance_block: remove dead stores
vp8_print_modes_and_motion_vectors: remove dead stores
This commit turns on the Hadamard transform based rate distortion
estimate for all block sizes in RTC coding mode. It conditionally
skips the rate distortion estimation if all zero block flag is set
on. No significant encoding speed change is observed. The
compression performance of speed -6 is improved by 1.7% over using
it only for block sizes of 32x32 and below.
Change-Id: I768145e6f05c737b05b5b5f1ee674e929532cafb
The threshold scaling factor was calculated wrong using partition
size "bsize". Thank Yaowu for pointing it out. It was fixed and no
speed change was seen.
Change-Id: If7a5564456f0f68d6957df3bd2d1876bbb8dfd27
The following functions use the count parameter to either loop or select
dedicated paths:
vp9_lpf_horizontal_16_c
vp9_lpf_horizontal_16_sse2
vp9_lpf_horizontal_16_avx2
vp9_lpf_horizontal_16_neon
vp9_highbd_lpf_horizontal_16_c
vp9_highbd_lpf_horizontal_16_sse2
Change-Id: I7abfd2cb30baa292b4ebe11c847968481103c037
This commit accounts for the transform block end of coefficient flag
cost in the RTC mode decision process. This allows a more precise
rate estimate. It also turns on the model to block sizes up to 32x32.
The test sequences shows about 3% - 5% speed penalty for speed -6.
The average compression performance improvement for speed -6 is
1.58% in PSNR. The compression gains for hard clips like jimredvga,
mmmoving, and tacomascmv at low bit-rate range are 1.8%, 2.1%, and
3.2%, respectively.
Change-Id: Ic2ae211888e25a93979eac56b274c6e5ebcc21fb
The vbp thresholds are set seperately for boosted/non-boosted
superblocks according to their segment_id. This way we don't
have to force the boosted blocks to split to 32x32.
Speed 6 RTC set borg test result showed some quality gains.
Overall PSNR: +0.199%; Avg PSNR: +0.245%; SSIM: +0.802%.
No speed change was observed.
Change-Id: I37c6643a3e2da59c4b7dc10ebe05abc8abf4026a
remove incorrect specializations in rtcd and update a configuration
check in partial_idct_test.cc
(cherry picked from commit 8845334097)
Change-Id: I20f551f38ce502092b476fb16d3ca0969dba56f0
To enable us to the scale-invariant motion estimation
code during mode selection, each of the reference
buffers is scaled to match the size of the frame
being encoded.
This fix ensures that a unit scaling factor is used in
this case rather than the one calculated assuming that
the reference frame is not scaled.
(cherry picked from commit 8d8d7bfde5)
Change-Id: Id9a5c85dad402f3a7cc7ea9f30f204edad080ebf
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
This commit allows the quantizer to compare the AC coefficients to
the quantization step size to determine if further multiplication
operations are needed. It makes the quantization process 20% faster
without coding statistics change.
Change-Id: I735aaf6a9c0874c82175bb565b20e131464db64a
For large partition blocks(block_size > 32x32), the variance
calculation is modified so that every 8x8 block's variance
is stored during the calculation, which is used in the
following transform skipping test. Also, the variance for
every tx block is calculated. The skipping test checks all tx
blocks in the partition, and sets the skip flag only if all tx
blocks are skippable. If the skip flag of Y plane is 1, a
quick evaluation is done on UV planes. If the current partition
block is skippable in YUV planes, the mode search checks fewer
inter modes and doesn't check intra modes.
The rtc set borg test(at speed 6) showed that:
Overall psnr: -0.527%; Avg psnr: -0.510%; ssim: -0.573%.
Average single-thread speedup on rtc set was 3.5%.
For 720p clips, more speedups were seen.
gipsrecmotion: 13%
gipsrestat: 12%
vidyo: 5 - 9%
dark: 15%
niklas: 6%
Change-Id: I8d8ebec0cb305f1de016516400bf007c3042666e
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
exclude files that only contain functions for non-high-bitdepth builds.
this removes some warnings related to missing prototypes
Change-Id: Ic6642998c46a7b808c6c53b2f9c34bcd4d037abe
This commit allows the encoder to check the eob per transform
block to decide how to compute the SATD rate cost. If the entire
block is quantized to zero, there is no need to add anything; if
only the DC coefficient is non-zero, add its absolute value;
otherwise, sum over the block. This reduces the CPU cycles spent
on vp9_satd_sse2 to one third.
Change-Id: I0d56044b793b286efc0875fafc0b8bf2d2047e32
When the estimated rate-distortion cost of skip coding mode is
lower than that of sending quantized coefficients, allow the
encoder to drop these coefficients. This improves the compression
performance of speed -6 by 0.268% and makes the encoding speed
slightly faster.
Change-Id: Idff2d7ba59f27ead33dd5a0e9f68746ed3c2ab68
This commit fixes the SSE2 version 8x8 Hadamard transform
alignment and makes it consistent with the C version.
Change-Id: I1304e5f97e0e5ef2d798fe38081609c39f5bfe74
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 makes the encoder to select between SATD/variance as
metric for mode decision. It also allows to account chroma
component costs for mode decision as well. The overall encoding
time increase as compared to variance based mode selection is about
15% for speed -6. The compression performance is on average 2.2%
better than variance based approach, with about 5% compression
performance gains for hard clips (e.g., jimredvga, nikas720p, and
mmmoving) at lower bit-rate range.
Change-Id: I4d04a31d36f4fcb3f5f491dacd6e7fe44cb9d815
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
webm_read_frame assumes that it won't be called once end of file
is reached. But for frame parallel mode that turns out to be not
true. this patch fixes that behavior by checking for EOS and
returning the appropriate value for subsequent calls.
Change-Id: Ie2fddbe00493a0f96c4172c67be1eb719f0fe8ed
Metrics on RTC set go down by ~1.5% on average.
Key frame encoding time goes down by factor of ~5.
Change-Id: Ia83acc55848613870e5ac6efe7f3d904d877febb
To enable us to the scale-invariant motion estimation
code during mode selection, each of the reference
buffers is scaled to match the size of the frame
being encoded.
This fix ensures that a unit scaling factor is used in
this case rather than the one calculated assuming that
the reference frame is not scaled.
Change-Id: Id9a5c85dad402f3a7cc7ea9f30f204edad080ebf
this matches the other includes and simplifies include paths in builds
from source
(cherry picked from commit 7999c07697)
Change-Id: I344902c84f688ef93c9f3a53e7c06c30db49d8d3
Set the GF group adaptive max Q compile flag to 1 by default.
This change has a quite big visual impact in some clips and also
contributes to tighter rate control.
For short test clips that have consistent content the impact is
quite small on metrics but for more varied long form clips there is
a drop in overal psnr but a sharp rise in average psnr caused by
greater expenditure on some easier sections and tighter rate clipping
in hard sections.
In chunck'ed encodes some of the effect will already be present due
to the independent rate control in each chunk but this change takes
the control down to a smaller scale.
yt hd +10.67%, - 3.77%, -1.56%
yt +9.654%, - 3.6%, - 1.82%
std hd +0.25%, -0.85%, -0.42%
derf +0.25%, - 1.1%. - 0.87%
Change-Id: Ibbc39b800d99d053939f4c6712d715124082843e
1. skip near if it is same as nearest
2. correct rounding for converting mv to fullpel position
3. update pred_mv_sad after new mv search.
Overall .1%~.25% compression gains on rtc set for speed 5, 6, 7, 8.
Change-Id: Ic300ca53f7da18073771f1bb993c58cde9deee89
Revised adjustment for rd based on source complexity.
Two cases:
1) Bias against low variance intra predictors
when the actual source variance is higher.
2) When the source variance is very low to give a slight
bias against predictors that might introduce false texture
or features.
The impact on metrics of this change across the test sets is
small and mixed.
derf -0.073%, -0.049%, -0.291%
std hd -0.093%, -0.1%, -0.557%
yt +0.186%, +0.04%, - 0.074%
ythd +0.625%, + 0.563%, +0.584%
Medium to strong psycho-visual improvements in some
problem clips.
This feature and intra weight on GF group length now
turned on by default.
Change-Id: Idefc8b633a7b7bc56c42dbe19f6b2f872d73851e
This experiment biases the rd decision based on the impact
a mode decision has on the relative spatial complexity of the
reconstruction vs the source.
The aim is to better retain a semblance of texture even if it
is slightly misaligned / wrong, rather than use a simple rd
measure that tends to favor use of a flat predictor if a perfect
match can't be found.
This improves the appearance of texture and visual quality
on specific test clips but is hidden under a flag and currently
off by default pending visual quality testing on a wider Yt set.
Change-Id: Idf6e754a8949bf39ed9d314c6f2daaa20c888aad
The joint_motion_search function alternates prediction
between two reference frames. In order to reuse existing
code, a pointer to the appropriate reference frame is
written into xd->plane[0].pre[0], that the motion
estimation code assumes points to the reference frame.
If this first reference frame was scaled then the
pointer was incorrectly being reset to point to the
unscaled reference frame rather than the scaled
version.
Change-Id: I76f73a8d8f4f15c1f3a5e7e08a35140cdb7886ab
While CONFIG_INTERNAL_STATS=1, PSNR is calculated while encoding.
The aligned width/height were used mistakenly in the calculation.
This patch fixed it, and used the orignal image width/height.
Change-Id: Iad5334f8693af761b71ebb78f2587db8a3404ecf
the TODO around CONFIG_SPATIAL_SVC has been resolved by changing the
CONFIG_* checks to use an ABI based check
Change-Id: If2638baf361b863186177a453beec9af9231e69e
this removes the CONFIG_* checks from public headers, but means
'--enable-experimental --enable-spatial-svc' builds will fail without a
local change to the ABI in vpx_encoder.h. this should be all right for
testing this experiment.
Change-Id: Ief55e7b9d1e8332cfce990275e04c29b30af0c4a
add explicit returns in cases where ASSERT_* can't be used due to the
function returning a value; retain the EXPECT_* for reporting purposes.
Change-Id: I1f514728537fee42a99277d3aba538e832d3b65b
Factor in segment#2 and skip blocks into the postencode estimated bits,
and increase somewhat the aggressiveness of the refresh.
PSNR/SSIM Metrics on RTC set go up by ~0.8/0.5%.
Change-Id: I5d4e7cb00a3aefb25d18c88b6b24118b72dc5d51
This commit makes the encoder to explicitly calculate the SAD
associated with the LAST_FRAME motion vector and compare it to
that of the GOLDEN_FRAME given by integral projection motion
estimation. It skips the expensive sub-pixel motion search over
GOLDEN_FRAME when the LAST_FRAME can provide fairly good motion
compensated prediction quality.
For dark720p speed -6 single thread goes from
33304 b/f, 40.070 dB, 18156 ms ->
33319 b/f, 40.061 dB, 17611 ms
Change-Id: I01bc94b9b598075567a392111046b97a9bc30efe
Because the call to vpx_codec_control at line 928 is now guarded by
!frame_parallel, 'corrupted' may not be set.
Change-Id: Id166bd8a8cdb5e5120fca1640011a3545f6e178a
Use force_split to constrain the partition selection.
This is used because in the top-down approach to variance partition,
a block size may be selected even though one of its subblocks may have
high variance.
In this patch the selection of the 64x64 block size will only
be allowed if the variance of all the 32x32 subblocks are also below the threshold.
Stil testing, but some visual improvement for areas near slow moving boundary
can be seen. Metrics for RTC set increase by about ~0.5%.
Change-Id: Iab3e7b19bf70f534236f7a43fd873895a2bb261d
Shut off all the metric checks for golden reference frame, if we
decide that it is unlikely to be selected for reference.
Change-Id: Ie457cc1fd43935584403b4982659aed80fb9909c
Move the scaling factor outside column projection. This avoids
repeated calculation of the same scaling factor. Profiling shows
that the percentage of vp9_int_pro_col_sse2 of overall cycles
goes from 2.29% down to 1.88%.
Change-Id: I5ac4e324ab2d7f33ba2de66dd2a12e04e04dfd66
1. remove duplicate initialization to mbmi->interp_filter.
2. move mv clamping into ref_frame loop instead of mode checking loop.
3. move the check if last frame is same as golden frame earlier to
avoid initialization of Golden reference related variables.
Change-Id: Idf2d05e19e94a24f69cc289687869fc71d2ff289
use \li to separate the list items contained in conditionals. this
avoids the encode page becoming a sub-item of decode; likely a problem
in <1.8.3.
+ fix encoder conditional, spelling error
+ correct encode page name to match decode 'Encoding'
Change-Id: I67890f52bed8e708bad63fb8819a074e0beff2ca
use \li to denote list items with \if.
fixes the following likely visible in <1.8.3:
usage.dox: warning: Invalid list item found
usage.dox: warning: End of list marker found without any preceding list items
Change-Id: I33c72799edf9f8866596ac8f79247050b8c75681
The compression performance of speed -5 is on average 12.6% better
than speed -6. At lower bit-rates, the gains are typically 20% or
more. For 2-thread encoding, the speed -5 takes about 1.6x time of
speed -6.
Change-Id: If7a73464a24d33e8f49b9533b51ec51c8da7fc80
The commit updates the comments in vp8cx.h to make it clear which
codec support each of codec control functions.
Change-Id: Ibf876e289d4325bbb61ce19311da60d384624c2f
Crash occured on very first key frame, because denoiser
temporal function was beng entered.
Updated denoiser unittest to set cpu_used from first frame,
and verified fix fixes the crash.
Change-Id: I3be1124b52846fbbe7248d2c3d6136e086c80bc1
Comments are updated to reflect that these controls apply to VP9 only,
thereby, to insure the document produced by doxygent to reflect the
same fact too.
Change-Id: Ic54c88ec066aa0ec4552d43dd4a7016e1f810f42
This commit uses a 6-point 1-step refine motion search in the
integral projection based full pixel motion estimation, to replace
the current 9-point search.
It reduces runtime cost of speed -6 on some noisy clips, e.g.,
dark720p single thread
33314 b/f, 40.076 dB, 18231 ms ->
33307 b/f, 40.067 dB, 17768 ms
The compression performance for rtc set remains unchanged.
Change-Id: I194ea5a9ce52e5a10baeee36338633adc22f764c
This commit changes to use single loop to evaluate all inter modes.
There is no impact on compression quality and speed, but allow future
experiment with the order of modes evaluated.
Change-Id: I71696ce1014cbe127e25e98710d835987f5ecc09
Added a skip_dc check. If skip_dc = 1, we could eliminate calling
of vp9_model_rd_from_var_lapndz(). This gave slight PSNR & SSIM
gain(<0.1%), and no speed change.
Change-Id: If5ca733366148c86b98e196a00cc890f50e9a3e5
Re-arrange the multiplication and right shift operations to avoid
integer overflow in choose_partitioning.
Change-Id: Ib4005cafb410a67c1960486471d75b6ebe38c4e0
This commit removes the pred_mv_sad comparison from rtc motion
search, given that a stronger comparison has been done at the
mode search level to eliminate unlikely selected reference frames.
Change-Id: I49b8d24b2174303066fd8eff2102c0648f2869df
This commit enables the rtc coding mode to run integral projection
based motion search for golden reference frame. It improves the
speed -6 compression performance by 1.1% on average, 3.46% for
jimred_vga, 6.46% for tacomascmvvga, and 0.5% for vidyo clips. The
speed -6 is about 6% slower.
Change-Id: I0fe402ad2edf0149d0349ad304ab9b2abdf0c804
Make the vp9_int_pro_motion_estimation() function return zero
motion vector if high bit depth is turned on, instead of removing
it from compiled codes.
Change-Id: Ia48f010eb590b2d517d5678c394110b326a1a95e
This patch accounts in the first pass stats for blocks that
while not coded as intra, are complex and have an intra error /
best error ratio below a threshold.
The modification shortens the GF arf interval for a particular
class of content that contains a lot of blocks matching the
above criteria. (In one short problem test sequence the average
interval dropped from about 14-15 to 10-11)
The change results in small net gains in metrics results for the
Yt(~0.2%) and yt-hd (~0.5%) sets and is approximately neutral
for the other test sets.
The change is currently shielded by a flag and off by default
pending verification that it does not cause other regressions
in tests on a wider YT test set.
Change-Id: I6b803daa6a4ac09a6f428fb3a18be1ecedd974b7
Only update the rd_thresh factors for modes sharing same reference
frame. This helps overall compression of 6 and 7 by .13% and .19%
respectively without any noticeable speed difference.
Change-Id: Idb3a3879512c5d7d0880034516079949290690c5
For non-SVC 1 pass CBR: make the GF update interval a multiple of the
cyclic refresh period, and use encoding stats to prevent GF update at certain times.
Change-Id: I4c44cacc2f70f1d27391a47644837e1eaa065017
Tx_totals counters weren't handled correctly in multi-thread
case, which caused the mismatch while encoding using threads > 1.
This patch fixed that.
Change-Id: Ice9b0386f57175fb92a0bdcd5042686a3106246a
The return value from vp9_compute_qdelta_by_rate, which is
a delta value for the quantizer, could never be 0 if
(qindex == rc->worst_quality).
This occurs because target_index was setup unconditionally
in the loop and yet the loop counter stopped at
(rc->worst_quality - 1).
Change-Id: I6b59cd9b5811ff33357e71cd7d814c5e53d291f2
Choose_partition uses only the last frame as reference frame in making
partition decision, this commit adds the check on how well Golden
frame with (0,0) predicts the current block, and uses GF(0,0) as
basis for partition decision if it produces better prediction.
The commit improves rtc speed 6 and 7 encoding by 0.14% and 0.19%
respectively.
Change-Id: I156acf925bd6e0b586d48155d1940d27270a3915
When golden reference frame is refreshed, the next frame has both
its last and golden reference frames point to the same reference
frame in real-time coding mode. Experiments suggest that using
two separate reference frames for frames right after golden refresh
frame does not provide further compression performance advantage.
This commit hence retains the current encoder implementation and
shuts off the mode search over golden reference frame in this case.
It makes the encoder run slightly faster at no coding performance
change.
Change-Id: I1561f7799253a10e675d05c63c1749fe9e85b472
Force 64x64 partitioning when a whole superblock is SEGMENT_LVL_SKIP. This
drops encode times of screens mostly at rest by 20%.
Change-Id: Ieba554b0b8a0c1679aae784a8bd11f038ab942c3
Adjustment previously only enabled in VBR mode.
This patch allows adjustment of min and max q for CBR
and adjustment of max q only for CQ mode.
Change-Id: Id5e583f3d50453cd544fc57249acacd946457482
While turning on "--aq_mode=3", the quantizers are updated by each
thread. Fixed the me consts initialization function to make sure
that the correct thread data are updated.
Change-Id: Ied27bb7bae76fc3fa2cda4f8c35ac0b46271bef4
While searching for the best mode in non-rd case, SSE of
a partition block is calculated and the transform size is set.
This patch rewrites the skip checking conditions based on
transform size instead of partition size to be more precise.
Small gains were seen in rtc set borg test (speed 6).
AVG PSNR: 0.087%, overall PSNR: 0.073%, SSIM: 0.146%.
No noticeable speed change.
Change-Id: I5603ca5339c784dfa02263f4005988ccd8c32f6e
It was tiny when it was orginally marked INLINE. Forcing this function
to be inlined prevents the compiler from inlining its much smaller
callers.
No measurable speed impact, 28320 byte smaller libvpx.a
Change-Id: I6bf4c917157d15cbadb3cd3e20a9e82d35dc7d6f
Visual Studio is exceptionally picky about this:
vp9_reconintra.c(900): warning C4113: 'void (__cdecl *)()' differs in
parameter lists from 'void (__cdecl *)(void)'
[.build-x86_64-win64-vs10\vpx.vcxproj]
Change-Id: I564c7415f4608fd962be8c699d6133a996b545f7
This saves an extra 64x64 variance calculation and replaces two
32x32 variance functions with sad functions. The compression
performance change is unnoticeable.
Change-Id: I6d33868695664ec73b56c42945162ae61c484856
Frame buffers are now allocated dynamically on-demand.
Entries in the reference frame map, cm->ref_frame_map,
may now be set to -1 (INVALID_IDX) to indicate that
there is not a valid reference buffer in that "slot".
All slots in the reference frame map are now initialized
to the empty state (-1) and each buffer is initialized
to have a reference count of 0.
Change-Id: Id1afe98de98db4ae8b2dfefed7889c3b28c68582
Use rectangular block size for integral projection motion estimation
if the the 64x64 block has over half block outside the frame. This
avoids the issue that the motion information of these blocks is
dominated by the extended pixels, instead of the pixels of interest.
Change-Id: I22f4d2bb7f6a20db9b3f5e2e5463a7f4b9d1b737
The rounding factor needs to be scaled down by a factor of 2.
Also, the quantized and dequantized coefficients are memset to 0
when dc quantizer is used.
Change-Id: Ifa68bab02addbf1b83d249c5b4cbd5cda796b1cf
A frame may be waiting for an out of border pixel from another
frame. A frame's row progress variable is set to -1 when start being decoded
and another frame may be waiting for -2 row pixel from this frame.
In this case, vp9_frameworker_wait will return directly and skip the waiting
which leads to tsan error between threads.
Change-Id: Id16604915fb598b823e34393f696e3aa46fb6422
Most of the current decoders use tile-based multithreading. Also
most of the current decoders need frame_parallel_decoding_mode
turned on to enable multithreaded decoding. tile-columns is
limited by resolution, so setting to max (6) is fine.
BUG=https://code.google.com/p/webm/issues/detail?id=963
Change-Id: I6e7ac3485d96bf0c69e06706cbb326dd38be0020
Instead using only a fixed threshold, this commit adapts the threshold
for color sensitivity decision to luma signal energy: chroma channel's
sse is at least 1/6 of that in luma for color sensitivity flag to be
set to active.
This recoups a large portion of the speed loss due to accounting for
chroma component costs in RTC mode decision.
Change-Id: Ie01f747f6037dba6a1d1ed3e10b71a0ef1abc42c
This patch fixed webm issue 962.
(https://code.google.com/p/webm/issues/detail?id=962)
The data races occurred when an encoder and a decoder were created
at the same time, and the function pointers were initialized twice.
Change-Id: I8851b753c4b4ad4767d6eea781b61f0ac9abb44b
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 fixes an issue in source frame border extension. It
causes certain frame resolution such as 640x480 to have a portion
of the right/bottom extension filled by zeros, which misleads
motion search and degrades transform coding performance when large
block size is used.
This fix improves the speed 2 compression performance of a few
yt sequence, typically ranging from 1% - 2%, up to 5% at median
to low bit-rate.
Change-Id: Id6b09a5695d9e7651c6dfbc2c6a72288b08af7fb
both the encode and decode perf tests require niklas_1280_720_30.yuv
broken since:
28eebf3 Merge "tests: add a shorter 720p test clip"
7839d03 tests: add a shorter 720p test clip
Change-Id: I51ebbf7261832e25d8f2c1da5c7df5c2e47f748e
+ add a helper function to reduce the duplication
this is a bit clearer when the environment variable is set, but the
directory is missing
Change-Id: I08f9b56122b5741bb40a5f795f7f82f5b49f1047
niklas_1280_720_30.y4m 60 frames @ 30fps
only a small number of frames are being used; this reduces the test data
download size in non-perf-test cases by >500M.
retain niklas_1280_720_30.yuv for encode+decode perf tests
Change-Id: I56b3433104acd462f952a9554280de5a3ec0b6d2
This commit applies one-step refinement search to the resulting
motion vector of the integral projectiion based motion estimation,
per 64x64 block. It improves the coding performance of speed -6.
pedestrian 1080p 500 kbps
51735 b/f, 36.794 dB, 16044 ms ->
51382 b/f, 36.793 dB, 16282 ms
cloud 1080p 500 kbps
24081 b/f, 37.988 dB, 14016 ms ->
23597 b/f, 38.076 dB, 12774 ms
vidyo1 720p 1000 kbps
16552 b/f, 40.514 dB, 8279 ms ->
16553 b/f, 40.543 dB, 8510 ms
The rtc set compression performance is improved by 0.5%.
Change-Id: I3d09bea2caf58b2a4f3b38aa26fffafcbe9a2c17
The intrinsic statement _mm_subs_epi16() should take immediate.
Feeding variable as its input argument will cause compile failure
in older version gcc.
Change-Id: I6a71efcc8d3b16b84715e0a9bcfa818494eea3f4
This commit modifies the hierarchical vector match patter. It
avoids repeated SAD computation at same points. The function
vp9_vector_sad_sse2 is called 12 times per 64x64 block, instead
of 15 times as before. The effective coverage remains the same.
Change-Id: I91ad9d27d40db8963c907d02af84e10702136994
In ssse3 functions, DEFINE_ARGS macro hard codes qcoeff and dqcoeff
to r3 and r4. If skip is 1, qcoeff and dqcoeff need to be loaded
from the stack, which doesn't work because of the above definitions.
Currently, skip=1 case is not used in the encoder. This patch fixed
the issue, so it can be turned on later.
Change-Id: I998d696b1a7a85dca2b3bcee790b21c21e039147
When GF group adaptive maxQ is enabled this patch accounts
somewhat for accumulated error in the rate control.
This improves accuracy quite a bit on many clips especially
when there is overshoot.
Examples when the overshoot and undershoot command line
parameters are set to 100:
Hall @ 1200 overshoot is reduced from 67-24%.
Akiyo @ 400 undershoot is reduced from 28%-15%.
Setting a lower value for undershoot or overshoot still
reduces the error further.
Impact on metrics is mixed with some gains in average psnr
but generally a little lower (e.g. 0.5%) on overall and ssim.
The GF group adaptation is still off by default in this patch.
Compared to with the head, enabling this mode now gives
big average psnr gains on the YT sets (e.g. YT_HD >11.2%),
a drop in overall PSNR (YT-HD 3.9%) and a smaller drop or
neutral for SSIM.
Change-Id: If4b32cd0740d3fb941317b374f9c2951954eee90
Target higher delta-qp for big blocks with zero motion,
and for segment#1: avoid 64x64 partition size and force 8x8 tx size.
Metrics on RTC set mostly positive: SSIM up by ~4%, PSRN by ~1.5%.
Doesn't seem to be any change in speed.
Change-Id: I1f68fa3c4f62dab3b90cc58041f05ebb048ae5ac
Modified the thresholds of deciding whether or not to skip
the transforms in model_rd_for_sb_y(). Used zbin[] instead
of dequant[] to be more precise. Also, modified the checking
coditions.
Rtc set borg test results (at speed 6) showed:
average PSNR gain: 0.138%, overall PSNR gain: 0.158%,
and SSIM gain: 0.177%.
The data rate test was modified slightly as suggested by
Marco.
Change-Id: Ieaf633ab77f4838cb3c45cf69065b29d55f8ae6c
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
There is a corner case that when a frame is corrupted, the following
inter frame decode worker will miss the previous failure. To solve
this problem, a need_resync flag needs to be added to master thread
to keep control of that.
Change-Id: Iea9309b2562e7b59a83dd6b720607410286c90a6
using this to control reallocation would miss a change if the function
were not called for every frame.
fixes potential memory corruption by the subsequent memset
Change-Id: I4c6bb6ab68803104fc824c7e27cc2f9b2cf53e33
use VP[89]_INSTANTIATE_TEST_CASE case when possible to disable the tests if
the codec is unavailable.
broken since:
be6aead Try again to merge branch 'frame-parallel' into master branch.
Change-Id: I8d81c5ba3b951f82be94bfaed6be194e4289baec
This commit prevent the encoder to update last_frame_type when a frame
is dropped in the encoder. Prior to this fix, if there is a dropped
frame immediatedly after a key frame, decoder would have the value of
last_frame_type as key frame, different from encoder as the dropped
frame in encoder would have updated the value to an inter frame. This
leads to different probability update in encoder and decoder, thereby
encoder/decoder mismatch.
This fixes issue #941
Change-Id: I27115224b138bec43ae3916c016574f5740822b0
Replaced a divide by 9 with 8, so some very small difference,
but otherwise no change in behavior.
Change-Id: I1079ae3c41e0789ff0bc6fa9940a238b6bca0f5b
Simple skin detection, from vp8; works reasonable on most of the
RTC clips, but could miss sometimes.
Added debug flag to write out skin map over source input.
Change-Id: I2caea7592f1c459047aac46627eeb24a94946464
This commit allows the encoder to properly account for the mode
cost in sub8x8 non-RD mode decision.
Change-Id: I2951960d20e37ed08e372ee0c7044935b2b9b899
Add the rate cost on inter prediction filter type to the overall
rate-distortion cost in vp9_pick_mode_inter.
Change-Id: I72c34017adf5220cadb3962694ee5404469fc673
This commit adds a heuristic rate cost of reference frame to the
non-RD mode decision. It improves the compression performance of
speed -6 by 0.31% and speed -5 by 0.69%.
Change-Id: If7f3b45519d49b2cb640bcb7316a254efc8be446
This enbale the encoder to set color space information for the input
video, so it is then coded in the output bitstream.
Change-Id: Ife03deab3c762425ccd27c4c190902c4d94a76f4
MODE_INFO struct was modified, and vp9_print_modes_and_motion_vectors()
didn't work anymore. This patch modified vp9_debugmodes.c so that
this function works again for debug usage.
Change-Id: I293fae0295235deb2529a460a274caf7c045ac1a
This is to avoid redo the same calculation repeatly, and also allow
easier adjustments for further experiments.
This commit shall have no effect on quality/compression.
Change-Id: I4460acf5c808ff5518da18d21e002c5da58af857
Note: This feature is still in development.
Add an option for the encoder to decide the resolution
at which to encode each frame.
Each KF/GF/ARF goup is tested to see if it would be
better encoded at a lower resolution. At present, each
KF/GF/ARF is coded first at full-size and if the coded
size exceeds a threshold (twice target data rate) at
the maximum active Q then the entire group is encoded
at lower resolution.
This feature is enabled in vpxenc by setting:
--resize-allowed=1
In addition, if the vpxenc command line also specifies
valid frame dimensions using:
--resize-width=XXXX & --resize_height=YYYY
then *all* frames will be encoded at this resolution.
Change-Id: I13f341e0a82512f9e84e144e0f3b5aed8a65402b
This commit fixes the sub block partition size used in
fill_mode_info_sb. Previous implementation effectively disabled
the rectangular block sizes. This commit resolved this issue.
Change-Id: Ic1c383ab0a9a2e7d59e85b388093f1f1f94d1e7f
This will fix the frame parallel decode hang on windows
due to not enough semaphores.
This will also make the frame parallel decode safer as
the number of frame buffers could only support maximum
8 threads.
Change-Id: Id9ef50692819dcbebbd74a0aabffbfb3f39a4309
This commit changes the value of highbitdepth flag to avoid conflict
with vp8 refresh_last_frame flag.
Change-Id: Idcff2cf44f0a200bd935b326f785c0cf32d7228a
The calculation of required extension used in HBD case was wrong due
to rounding for UV when y dimension is odd. This commit replace the
computation with correct version.
This fixes a crash caused by writting beyond buffer boundary.
Change-Id: Ic7c9afeb7388cd1341ec4974a611dacfb74ac6b6
This commit makes the ZEROMV mode first in the search order to
ensure that the zero mv is always checked in the RTC coding mode.
It improves the average speed -6 compression performance by 0.3%
in both PSNR and SSIM at no visible speed change.
Change-Id: I465a7e59f4e20cd84fee3f02ced6f98036945949
This reverts commit a6715a7558.
Removes a duplicate entry; this was previously added by:
14e37cf Add help info for --enable-vp9-highbitdepth
Change-Id: I61408e782232821ef6ed84775b5c79d172ba7f2d
cm->frame_bufs[].idx values were made consistent in:
61c5e94 Use -1 consistently as invalid buffer idx
update the initialization in swap_frame_buffers() to match.
additionally:
- remove some shadowed variables in the former and marked them volatile
Change-Id: Ie3f9636c405bd822112bb56bd22d28024ae98909
The high bit depth build failed while building for 32bit target.
The bugs were in vp9_highbd_subpel_variance.asm and
vp9_highbd_sad4d_sse2.asm functions. This patch fixed the bugs,
and made 32bit build work.
Change-Id: Idc8e5e1b7965bb70d4afba140c6583c5d9666b75
there are no known issues since:
10d5e09 Fix issues in 32bit PIC enabled build
related issues: #808, #924
Change-Id: I80454f95fe6b4ce630fdd434d740ce8b0d42951b
The current file's directory, ".", is treated much more literally
when building libvpx examples with Xcode than it is with make, and
clang cannot find common include files included via "./" when those
files actually reside one directory up in the tree.
Change-Id: I5f66a026282e35d80248ca4052ebb882b859172e
On rtc set:
speed 7 quality improves about 0.5%
speed 8 quality improves about 1.0%
Encoding time for speed 7 changes from 67804ms to 65889ms
Encoding time for speed 8 changes from 58659ms to 56808ms
Change-Id: Iabcfb53012fc1b9f3326cdbc167e5758b8c7ad30
After syncing the frame worker thread, avaiable thread count should
increase by 1 even the worker thread does not have displayable frame
to output.
Change-Id: I9eeb87720fed82dfe38555286833ff88e8a8e746
Reuse the yv12_mb array to fetch the buffer pointers/strides
corresponding to the current reference frame.
Change-Id: I5276b7494158b2cccef15213be2dc189e9036851
This commit allows the encoder to account for additional chroma
plane costs in the mode decision process, if the current block
potentially contains significant color change. It improves the
visual quality at very low bit-rates.
The compression performance of dark720p is improved by 12.39% in
speed 6. For jimred at 150 kbps, the PSNR of V component (red)
increased by 0.2 dB, at the expense of about 5% increase in
encoding time. Note that for sequences where the chroma components
are fairly consistent, the encoding time increase is negligible.
On average the rtc set compression performance is improved by
1.172% in PSNR and 1.920% in SSIM.
Change-Id: Ia55b24ef23a25304f7ec9958fbf07fd6e658505c
This patch continues the work to remove frame_parallel_decoding_mode
requirement in VP9 multi-threaded tile decoder. In order to do that,
the frame counts associated to each thread need to be accumulated
together after the frame is decoded.
Change-Id: Idba1a756cedfed3c154aef52ed82c8da3bbf9e0c
The original implementation had the following comment:
// Ignore mv costing if mvsadcost is NULL
However the current implementation does not allow for this.
If x exists then nmvsadcost must not be null.
This removes the only warning from -Wpointer-bool-conversion
https://code.google.com/p/webm/issues/detail?id=894
Change-Id: I1a2cee340d7972d41e1bbbe1ec8dfbe917667085
The current multi-threaded tile decoder requires that the videoes
are encoded with frame_parallel_decoding_mode = 1. This requirement
is not necessary, and is better to be removed. This patch includes
the first part of the work.
Change-Id: Ic7695fb3cfe13f9022582c9f0edd2aa6e2e36d28
In vp8_sub_pixel_variance8x8_neon the temp2 buffer is only initialized
to kHeight8 * kWidth8. However, in the case that xoffset != 0 and
yoffset == 0, var_filter_block2d_bil_w8 is called with output_width
kHeight8PlusOne.
Thanks to cmugurel for diagnosing and yulius for the patch.
Change-Id: Ib71ffd96ffad963c92b8b7ca23f303942785b8e0
https://code.google.com/p/webrtc/issues/detail?id=4190
Apple ships version 0.98 of nasm through at least XCode 6. It is
incompatible with the assembly in libvpx.
https://code.google.com/p/webm/issues/detail?id=772
Change-Id: I33245a76f50a8224fe6fafa3cce9991f953fdcc8
1. Adjusted the threshold for coef update computation based on counts
of tx used, avoid coef update computation when count is low (<20)
2. Move sf->lpf_pick = LPF_PICK_MINIMAL_LPF to speed 8.
Change-Id: I02b44309e40fcdbf135c7934ae067a3f42502d30
In frame parallel decode, libvpx decoder decodes several frames on all
cpus in parallel fashion. If not being flushed, it will only return frame
when all the cpus are busy. If getting flushed, it will return all the
frames in the decoder. Compare with current serial decode mode in which
libvpx decoder is idle between decode calls, libvpx decoder is busy
between decode calls.
Current frame parallel decode will only speed up the decoding for frame
parallel encoded videos. For non frame parallel encoded videos, frame
parallel decode is slower than serial decode due to lack of loopfilter
worker thread.
There are still some known issues that need to be addressed. For example:
decode frame parallel videos with segmentation enabled is not right sometimes.
* frame-parallel:
Add error handling for frame parallel decode and unit test for that.
Fix a bug in frame parallel decode and add a unit test for that.
Add two test vectors to test frame parallel decode.
Add key frame seeking to webmdec and webm_video_source.
Implement frame parallel decode for VP9.
Increase the thread test range to cover 5, 6, 7, 8 threads.
Fix a bug in adding frame parallel unit test.
Add VP9 frame-parallel unit test.
Manually pick "Make the api behavior conform to api spec." from master branch.
Move vp9_dec_build_inter_predictors_* to decoder folder.
Add segmentation map array for current and last frame segmentation.
Include the right header for VP9 worker thread.
Move vp9_thread.* to common.
ctrl_get_reference does not need user_priv.
Seperate the frame buffers from VP9 encoder/decoder structure.
Revert "Revert "Revert "Revert 3 patches from Hangyu to get Chrome to build:"""
Conflicts:
test/codec_factory.h
test/decode_test_driver.cc
test/decode_test_driver.h
test/invalid_file_test.cc
test/test-data.sha1
test/test.mk
test/test_vectors.cc
vp8/vp8_dx_iface.c
vp9/common/vp9_alloccommon.c
vp9/common/vp9_entropymode.c
vp9/common/vp9_loopfilter_thread.c
vp9/common/vp9_loopfilter_thread.h
vp9/common/vp9_mvref_common.c
vp9/common/vp9_onyxc_int.h
vp9/common/vp9_reconinter.c
vp9/decoder/vp9_decodeframe.c
vp9/decoder/vp9_decodeframe.h
vp9/decoder/vp9_decodemv.c
vp9/decoder/vp9_decoder.c
vp9/decoder/vp9_decoder.h
vp9/encoder/vp9_encoder.c
vp9/encoder/vp9_pickmode.c
vp9/encoder/vp9_rdopt.c
vp9/vp9_cx_iface.c
vp9/vp9_dx_iface.c
This reverts commit a18da9760a.
Change-Id: I361442ffec1586d036ea2e0ee97ce4f077585f02
+ nearest for consistency
near is a reserved word in windows builds so using it as a parameter
name may cause build failures with some configurations
Change-Id: Iddf1d4ecdb39843f14e95dbfd9dca55f07f81403
1. move the check of search method of USE_TX_8X8 up one level to
avoid operations of build_tree_distributions()
2. count tx used and avoid computaton for coef udpate when one size
is not used at all.
Change-Id: Ia3e54a2588aa531c41377a1bfaa64385d04a592c
The previous patch "Fix issues in 32bit PIC enabled build" fixed
the x86inc.asm for macho32. Now we can enable use_x86inc while
building libvpx for 32bit pic enabled Darwin target, which makes
the encoder a lot faster(>2X) in this case by turning on the
existing optimizations.
Change-Id: I5f5c7add428d73f50c935c48d0a70aed2b1eb7af
gives a better summary of what is enabled / disabled outside of the
automatic toolchain options.
fixes issue #936
Change-Id: I1bf27593a5512713aab1473cb606c58cf3084d62
1. reduce the size of temporaray arrays on stack
2. avoid build_tree_distribution for tx size that is not used at all.
Change-Id: I0f8d7124e16a3789d3c15ad24cf02c1c12789e2c
This patch was to fix issue 924:
https://code.google.com/p/webm/issues/detail?id=924
The SECTION_RODATA macro was modified to support macho32 format.
The sub-pixel functions were modified to pass in 2 more parameters
to handle the global offsets for PIC build.
Change-Id: I3bfcd336bcae945edf300bca4ab40376a2628cd4
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
The current method doesn't work with Xcode 4 and up, since they no
longer have a $DEVELOPER_DIR/SDKs directory. Using xcrun and xcodebuild
works all the way back to Xcode 3 on OS X 10.6 Snow Leopard, if not
earlier.
Change-Id: I7126f2fb4a8f1d6e46f921e70bbd090f00ce3d36
Floating point is used in vp9_convert_qindex_to_q(), so sometime unit
test ActiveMapTest would cause run time error without properly call
to clear_system_state to reset register status.
Change-Id: I181e9395148c44a6ca8b97d6e109bd4a152143c6
Add distortion threshold condition to refresh state of a coding block,
and allow for qp adjustment also for some intra modes and non-zero motion modes.
Also some code cleanup (remove unused variables/code).
Change-Id: I735fa2b28bc64f60e0323976b82510577b074203
Currently disabled by default: enabled using
#define GROUP_ADAPTIVE_MAXQ
In this patch the active max Q is adjusted for each GF
group based on the vbr bit allocation and raw first pass
group error.
This will tend to give a lower q for easy sections
and a higher value for very hard sections. As such it is
expected to improve quality in some of the easier
sections where quality issues have been reported.
This change tends to hurt overall psnr but help
average psnr. SSIM also shows a small gain.
Average results for derf, yt, std-hd and yt-hd test sets were
as follows (%change for average psnr, overal psnr and ssim):-
derf +0.291, - 0.252, -0.021
yt +6.466, -1.436, +0.552
std-hd +0.490, +0.014, +0.380
yt-hd +5.565, - 1.573, +0.099
Change-Id: Icc015499cebbf2a45054a05e8e31f3dfb43f944a
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 result should have both bits set; previously this was converted from
webp incorrectly and resulted in a boolean check...
Change-Id: I2a7c7f2b491945f3a536ab4fca02247eccc892b8
This commit replaces an integer divide with a table-lookup. It is
to improve decoding speed, and at the same time, to reduce possible
complications with a bug in AMD Family 12h processors:
"665 Integer Divide Instruction May Cause Unpredictable Behavior"
Change-Id: I678b707a538798a923850bac467e66e847e6def7
In frame parallel decode, libvpx decoder decodes several frames on all
cpus in parallel fashion. If not being flushed, it will only return frame
when all the cpus are busy. If getting flushed, it will return all the
frames in the decoder. Compare with current serial decode mode in which
libvpx decoder is idle between decode calls, libvpx decoder is busy
between decode calls. VP9 frame parallel decode is >30% faster than serial
decode with tile parallel threading which will makes devices play 1080P
VP9 videos more easily.
* frame-parallel:
Add error handling for frame parallel decode and unit test for that.
Fix a bug in frame parallel decode and add a unit test for that.
Add two test vectors to test frame parallel decode.
Add key frame seeking to webmdec and webm_video_source.
Implement frame parallel decode for VP9.
Increase the thread test range to cover 5, 6, 7, 8 threads.
Fix a bug in adding frame parallel unit test.
Add VP9 frame-parallel unit test.
Manually pick "Make the api behavior conform to api spec." from master branch.
Move vp9_dec_build_inter_predictors_* to decoder folder.
Add segmentation map array for current and last frame segmentation.
Include the right header for VP9 worker thread.
Move vp9_thread.* to common.
ctrl_get_reference does not need user_priv.
Seperate the frame buffers from VP9 encoder/decoder structure.
Revert "Revert "Revert "Revert 3 patches from Hangyu to get Chrome to build:"""
Conflicts:
test/codec_factory.h
test/decode_test_driver.cc
test/decode_test_driver.h
test/invalid_file_test.cc
test/test-data.sha1
test/test.mk
test/test_vectors.cc
vp8/vp8_dx_iface.c
vp9/common/vp9_alloccommon.c
vp9/common/vp9_entropymode.c
vp9/common/vp9_loopfilter_thread.c
vp9/common/vp9_loopfilter_thread.h
vp9/common/vp9_mvref_common.c
vp9/common/vp9_onyxc_int.h
vp9/common/vp9_reconinter.c
vp9/decoder/vp9_decodeframe.c
vp9/decoder/vp9_decodeframe.h
vp9/decoder/vp9_decodemv.c
vp9/decoder/vp9_decoder.c
vp9/decoder/vp9_decoder.h
vp9/encoder/vp9_encoder.c
vp9/encoder/vp9_pickmode.c
vp9/encoder/vp9_rdopt.c
vp9/vp9_cx_iface.c
vp9/vp9_dx_iface.c
Change-Id: Ib92eb35851c172d0624970e312ed515054e5ca64
For low spatial resolutions: bias partittion selection to smaller block sizes,
and base the variance computation on 4x4 down-sampling.
Also move the threshold computations into the choose_partitioning,
so they are computed once for each sb block.
On low-res clips (RTC_derf) PSNR/SSIMetrics increase by about 4-5%.
No change for resolutions above CIF.
Change-Id: I93f8ff742c8044786977bb6e31dcf8efda6dd1b0
Just before a forced key frame we often get a foreshortened
arf/gf group. In such a case, we do not want to update
rc->last_boosted_qindex, which is used to define the Q range
for the forced key frame itself.
This gives a small average metrics gain for the YT and YT-HD sets
(eg. YT SSIM +0.141%).
Change-Id: Ie06698bc4f249e87183b8f8fb27ff8f3fde216d9
The comparison of address in the condition is not necessary, since
they will constantly be non-null.
Change-Id: Id0b0075283f5af65215d5761a8160a4cb2a15c9b
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
1. Added row-based loopfilter in encoder;
2. Moved common multi-threaded loopfilter functions from decoder
to common;
3. Merged multi-threaded loopfilter code, and made encoder/
decoder call same function to reduce code duplication.
Encoder tests showed that 1% - 2% speedup was seen for good-quality
2-pass mode(at speed 3); 1% - 3% speedup using 2 threads and 4% - 6%
speedup using 4 threads were seen for real-time mode(at speed 7).
Change-Id: I8a4ac51c2ad9bab9fa7b864e90743931c53ec1c4
This commit fixes a bug in denoiser reference frame buffer swap,
which disables frame buffer update.
Change-Id: I39a9427180fd18f9692602064ad821f7af4714c0
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 some platforms, such as 32bit Windows and 32bit Mac, the allocated
memory isn't aligned automatically. The thread data is aligned to
ensure the correct access in SIMD code.
Change-Id: I1108c145fe982ddbd3d9324952758297120e4806
This commit added coments for the following encoder controls:
VP9E_SET_LOSSLESS
VP9E_SET_TILE_COLUMNS
VP9E_SET_TILE_ROWS
VP9E_SET_FRAME_PARALLEL_DECODING
VP9E_SET_AQ_MODE
Change-Id: I2f75afd9cce01394f202b8e25f36bf763be0ddeb
This commit adds encoder side control for vp9 to set color space info
in the output compressed bitstream.
It also amends the "vp9_encoder_params_get_to_decoder" test to verify
the correct color space information is passed from the encoder end to
decoder end.
Change-Id: Ibf5fba2edcb2a8dc37557f6fae5c7816efa52650
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
Don't put small empty frame in front of a key frame. We will
put key frame flag in webm container if there's a visible key
frame. But there will be decoding error when we seek to here
if we put the small empty frame, which will be inter frame,
in front of it.
Change-Id: Id50c2c1fd31da0405ff6faa7375cc2f49c55402d
This commit added a field to vpx_image_t for indicating color space,
the field is also added to YUV_BUFFER_CONFIG. This allows the color
space information pass through the decoder from input stream to the
output buffer.
The commit also updated compare_img() function with added verification
of matching color space to ensure the color space information to be
correctly passed from encode to decoder in compressed vp9 streams.
Change-Id: I412776ec83defd8a09d76759aeb057b8fa690371
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
Replaced "color space" with "color format" in comments where color
sampling format is concerned, so to differentiate from the concept
defined in COLOR_SPACE.
Change-Id: I8c935034c166b24307a99352dab1686531276bb8
This commit refactors the motion compensated reference block fetch
process in denoiser. It skips the stage that generates motion
compensated reference block if denoiser decides to use copy block
mode. For high motion clips, this could speed up the denoising
process by about 10%.
Change-Id: I8ef4fa5fe766a8c4529119b9ec01faefb3d4ef53
Use frame buffer pointer swap instead of memcpy when possible.
These two CLs make the denoiser when running on vidyo1 720p at
speed -6 over 10% faster.
Change-Id: I64fe8a2422cafca6787a50c7f4dfb961191c0a9d
These two parameters are used to control the denoiser cut-off
thresholds. They should be properly initialized when starting
mode search of a given block.
Change-Id: Iba8a25487026a0dbe0d350c347d7e4e4e237b637
When qdiff is larger, the sad/variance threshold should also be
higher which indicates a more aggressive action on MFQE.
Change-Id: I44c5c93572805458d4f87fdc7619cc9d8a522185
The vp9_denoiser_free() function will internally check if the
buffer pointers are NULL. This commit makes the encoder always
call vp9_denoiser_free() after finishing encoding. It protects the
case where noise_sensitivity_level is changed during encoding
process and happen to be turned off towards the end of sequence,
which could result memory space allocated to denoiser not being
released.
Change-Id: Ie20dc2f2e6e5fb6333fbab3356bc153978a6a0f8
Use the correct frame size and stride value for chroma components
when setting the initial values. These control parameters are
assigned when the denoiser buffer was allocated and initialized.
Change-Id: Ia6318194c7738aff540bcbd34f77f0dac46221a1
Allocate the frame buffer allocation for denoiser once during the
encoder initialization. This avoids allocating frame buffer
multiple times and overwriting the buffer pointer without proper
releasing.
Change-Id: I9b3baa6283449d86fd164534d344c036bb035700
When testing frame sse to choose a loop filter value and
when checking ambient error in kf Q selection, use 64 bit
values for accumulating the sse, to avoid risk of overflow
for large image formats.
Change-Id: I03765d16c843d0ade61a45b0cd46312472697e57
Separate functions and rename files. This will make it easier to disable
some functions later to help work around a compiler issue in chromium.
Change-Id: I7f30e109f77c4cd22e2eda7bd006672f090c1dc5
This makes the inter_mode counts update consistent with other symbols.
Also, forward updates should work corerctly now.
Change-Id: Id98be26fd08875162e644bb8f1de6f0918f85396
The denoiser sensitivity level should be set to 1 starting from
key frame. The internal function of denoiser should make the
temporal denoising operations cut off in key frame coding.
Change-Id: Id3e704a73e98e4ea801284a2cbbab2ea9c371d23
With "show_existing_frame" frames:
Minimum data size for profile 0 and 1 is 1 byte (8bits)
Minimum data size for profile 2 and 3 is 2 bytes (9bits)
Otherwise:
Minimum data size is 8 bytes.
This resolves the VP9 failure in fuzzing test build #56.
Change-Id: I146d9d37688f535dd68d24aacc76d464ccffdf04
By using weighted averaging in the calculation of the frames to be
displayed, we get an average gain of more than 1 db for key frames
whose base qp are 20 higher than non-key frames.
Change-Id: I7bcb2e7b9c6420ea3f73f33204d18b072dffd17c
This commit fixes the buffer alignment control in denoised video
output function. The encoder is now able to properly store the
denoised input video into provided file when enabled.
Change-Id: I258e272c8d4a9b52592e16d6d09976c6f5c21728
Use mbmi->segment_id directly in vp9_pick_inter_mode. The value is
set outside this function, hence no need to assign it again.
Change-Id: I3d63cdd2e4fadf62ccdefada638b00d979eb3741
Check if block size is below 8x8 for rectangular block coding. It
is added to support 4x8 and 8x4 block coding for RTC mode.
Change-Id: I760b328f45b98ae48adc45ed5a39fb643cd8aebd
This commit simplifies the reference motion vector part for sub8x8
block coding in RTC mode and reduces the required local variables.
Change-Id: I470d1482092563b68af22404dc1f497e7457b0a8
VP9FrameSizeTestsLarge.OneByOneVideo has been causing a failure in
jenkins libvpx__unit_tests-valgrind_long for "using of uninitialized
memory", the root cause was that the input image for this test was
not initialized with proper size, therefore plan U and V were not
initialized at all.
This commit fixes the size initialization, and resolves the issue.
Change-Id: Ic4dd1542b7bb0cb260a1e0aeeb505db21ae5edc8
This commit enables sub8x8 inter block coding for RTC mode. The
use of sub8x8 blocks can be turned on by allowing
choose_partitioning function to select 4x4/4x8/8x4 block sizes.
Change-Id: Ifbf1fb3888fe4c094fc85158ac3aa89867d8494a
Properly set the corresponding scaling factor of the reference
frame in the non-RD mode decision process. This allows the mode
search process to account for the scaled reference frame when
selecting coding mode.
Change-Id: I9d41bff6931c98e5a82b413e37ac5e6e14b93b23
Local variables used at the setjmp() site need to be marked volatile.
Relevant excerpt from the 'man longjmp':
===============
The values of automatic variables are unspecified after a call to
longjmp() if they meet all the following criteria:
· they are local to the function that made the corresponding setjmp(3) call;
· their values are changed between the calls to setjmp(3) and longjmp(); and
· they are not declared as volatile.
===============
Change-Id: I093e6eeeedbf5f781d202248ca701ba2c29d3064
This allows us to track decode speed for new encodes so that we catch
problems like an encode change that makes decode really slow.
Change-Id: I7210196415c4e53d455e9c81246d9fb324913a06
Encode the files with 1, 2, and 4 threads.
Explicitly turn on error resilient and frame parallel
decoding and turn off altref frames.
Change-Id: I02b66f72b7d35c666c3ba685b33015508e440209
The unit tests for VP9 multi-threaded encoder are added, which
carry out tests for all three modes(i.e. kTwoPassGood, kOnePassGood
and kRealTime), and speeds ranging from 1 to 8. A 1280x720 test
clip is used, which is encoded into multiple tiles. The number of
threads is num_of_tiles.
Change-Id: I04419eeca145ad841c9c527603668239a82e7fbd
This commit adds a guard condition to the intra mode test skip
control in RTC coding mode. If all inter modes are skipped, force
the encoder to check intra mode. It avoids situations where the
encoder processes without properly assigning required mode
information.
Change-Id: Ibb349fee997d6584ce901d08b06e8df3ca9c01b1
Initial patch to remove get_zbin_mode_boost() and
cpi->zbin_mode_boost.
For now sets a dummy value of 0 for zbin extra pending
a further clean up patch.
Change-Id: I64a1e1eca2d39baa8ffb0871b515a0be05c9a6af
this function may return an error if no frame is available; --keep-going
is meant to test decoder resilience, so simply warn in this case.
Change-Id: I6e6aed3e78eca21cca80d7d8a06a1a244685ba29
The alternate reference frame is disabled in non-RD mode. No need
to keep the related entries in the THR_MODES array.
Change-Id: I53386f4bb1c6284f582801f27246c5edf55bc24b
In RTC coding mode, the alternate reference frame modes and compound
inter prediction modes are disabled. This commit reworks the
related mode search threshold update process to skip interacting
with these coding modes. It provides about 1.5% speed-up for speed
-6 on average.
vidyo1
16551 b/f, 40.451 dB, 6261 ms -> 16550 b/f, 40.459 dB, 6190 ms
nik720p
33316 b/f, 38.795 dB, 6335 ms -> 33310 b/f, 38.798 dB, 6237 ms
mmmoving
33265 b/f, 41.055 dB, 7176 ms -> 33267 b/f, 41.064 dB, 7084 ms
dark720
33329 b/f, 39.729 dB, 11235 ms -> 33331 b/f, 39.733 dB, 10731 ms
Change-Id: If2a4090a371cd28f579be219c013b972d7d9b97f
This commit removes undefined value options of cpu-used for VP9 and
changed vpxenc prompt to reflect the usable range of [-8,8]
Change-Id: Ib80fef3dbb6ec9aabac45ed13e8ab6fbaf94f55e
Use a temporary variable to store the transform size associated
with the best intra mode and restore the mode_info if the overall
best mode is intra mode.
Change-Id: I2606e0061ad32f91b095462902b1eb734b128eea
The encoder initialization is called in EncodeFrame(). Therefore,
in the unit tests, the set control is done when video->frame() is 1.
This didn't cause problem since current tests mainly test lag_frame
> 0 case, or no encoding option that needs to allocate memory before
1st frame is used. If use lag_frame = 0 and encoding multiple tiles,
the unit tests crash. The issue is fixed by doing the initialization
before encoding frames.
Change-Id: I43102048f88448bcf27e9c60e0ec06c176b02e5c
Only for the rectangle blocks larger than 16X16, SAD and Variance are
still based on the internal square blocks.
Change-Id: I3754da1b0254147313f86a0140dbf4f980f06a5a
The mode_info array was unnecessarily reset to zero every frame
when error resilient mode turned on, given that the mode info
values per block will be assigned during mode search stage.
This commit removes this reset operation. It reduces the runtime
cost on memset operation to 1/3. The overall speed -6 runtime is
reduced by 2%.
Change-Id: I32ecb73338d8995cc0c5147de09357364f13d45b
This commit explicitly set the second reference frame type to be
NONE in key frame coding mode. This fixes a subtle dependency of
reference motion vector used by next inter frame on mode_info
reset before key frame coding.
Change-Id: I5ff0359753fdc9992b0bfe889490f7a32d7d5f6a
These were established for compatibility. Make sure to use them.
Most frequently they manifest as issues on Visual Studio builds.
Change-Id: I39d764d2eb341b999d7a6132cb44b2acfc511160
Export vpx_codec_enc_init_multi_ver so the vp8 multi res encoder example
can see it when building shared.
Change-Id: Ic5222b1b6d949f39c7e50c3bc58fb76bece2a3f1
Delete vp9_dc_only_idct_add_neon.c
The function was merged with vp9_short_idct4x4_1_add (later
vp9_idct4x4_1_add) in d2de1ca and should have been deleted then.
Change-Id: Ie58ba3dd9dc7330a8f1238dd7dd71c9ed4639b94
Signed-off-by: James Yu <james.yu@linaro.org>
Where there is very subtle motion, especially when combined
with low spatial complexity, the codec sometimes fails to quickly
pick up the ambient motion field.
Once it has been established though the field propagates well using
Nearest and Near MV.
This patch looks specifically at the case where the Nearest and Near
have not been established as non zero vectors and in this case
discounts the cost of searching for a new vector in the rd code.
This will almost certainly have some implications in terms of encode
speed but it should be possible to mitigate the impact in a subsequent
using first pass stats and the local spatial complexity.
Average results for test sets approximately neutral.
Change-Id: I44a29e20f11f7ab10f8c93ffbdc50183d9801524
Change 72141 introduced a new use of vp9_avg_4x4.
This call needs to switch to using vp9_highbd_avg_4x4
when performing high bitdepth encodes.
Change-Id: I6a8ba4b62f8a75d0a917b365a55245e2f0438ea1
When multiple intra modes are tested, the previous mode info
update process may overwrite the selected best intra mode and make
the final selection use an inter mode. This commit fixes this
issue by moving the mode_info reset outside the intra mode search
loop.
Change-Id: I15ed4288a6b3cb0832104a5e6d5d9a25cd1a5b2b
If vp9_pick_inter_mode works properly, it should at least check
one coding mode and hence get best_tx_size assigned a valid value.
There is no need to initialize best_tx_size with a legitimate
value before starting the mode search.
Change-Id: Ic0496cd89672ea9c2c512a9bd1da952190af9cba
Make the variable reduction_fac log2 based and explicitly use
right shift when computing intra_cost_penalty.
Change-Id: I208f1fb879a02debb3b3fc64f9fd06260dcf1c86
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>
Fails to compile. Bad calls to vp9_alloc_frame_buffer
and vp9_realloc_frame_buffer in postproc.c
This reverts commit 399823b6f5.
Change-Id: I29f0e173f8e185d3a303cfdb17813e1eccb51e3a
Allows override of default target list. Also added missing usage info
for --extra-configure-args, and removed last vestiges of armv6 support.
Change-Id: Ic0f14fffa0cbaea1bed371d38ff65e035bbe3273
Add support for setting byte alignment on the Y, U, and V plane of the
reference buffers. The byte alignment must be a power of 2, from 32 to
1024. A value of 0 sets legacy alignment.
Change-Id: I7c1399622f7aa68e123646369216b32047dda73d
INLINE is used quite widely in vp9, this change improves performance
1-2% on most modern platforms.
Change-Id: I8a9974aab89fa588ea4923cc7eaf6199e344a528
the entire module is wrapped in CONFIG_VP9_POSTPROC which is forcibly
enabled with CONFIG_INTERNAL_STATS
+ a similar change in vp9_alloccommon.c
Change-Id: I374993297a9fba5bef2f0b71f984eba42f0995a3
set LIBVPX_RAND with --enable-vp9-postproc, previously only the vp8
config was checked. this fixes the build with --disable-postproc.
Change-Id: Ia61baded6aa0e44d6443ae4a3c85915f1054f053
Assembly tests should clear system state, as we have no
expectation of proper system state in between test runs..
Change-Id: I0f591996c1f17ef2a5a8572a6b445f757223a144
This commit fixes a bug in the PICK_MODE_CONTEXT index for
horizontal partition case. The compression performance change
is less than 0.01% level, since most blocks are selected to
use square block size in RTC coding mode.
Change-Id: I67effc18ae8795fccdd82a55f4efc609fa5cb3e1
For key frame under variance source partition: 4x4 prediction blocks
may be selected when variance of 8x8 block is very high (threshold is set fairly high for now).
Testing on some RTC clips shows this helps to reduce some ringing artifacts on key frame.
Encoded key frame size increases about ~10%. Key frame PSNR increases about ~0.1-0.2dB.
Change-Id: I56e203fac32ea6ef69897fb3ea269c59cb50d174
This commit explicitly uses the bit shift operation instead of
division for computing block variance.
Change-Id: Id19c0ff27dd1d1ae4aceee6657e1aad0d406bd74
If decoding starts with intra-only frame, there is a possibility
of using uninitialized entropy context, what leads to undefined
behavior.
Change-Id: Icbb64b5b1bd1e5de2a4bfa2884e56bc0a20840af
The 8x8 DCT uses a fast version whenever possible.
There was a mistake in the checking code which
meant sometimes the fast version was used when it
was not safe to do so.
Change-Id: I154c84c9e2d836764768a11082947ca30f4b5ab7
(cherry picked from commit fd05fb0c21)
This commit refactors the choose_partitioning function. It removes
redundant memset calls and makes the encoder to calculate
variance value per block only when it is needed. It reduces the
average runtime cost of choose_partitioning by 60%. Overall it
reduces speed -6 runtime by 2-5%.
Change-Id: I951922c50d901d0fff77a3bafc45992179bacef9
It is the first version of MFQE in VP9. There are a few TODOs included
in this version.
Usage: Add flag --enable-vp9-postproc to config the project.
In decoder, use flag --mfqe in the command line to enable
MFQE in postproc.
Note: Need to have key frame with low quality to see the effect of this
new patch. In my experiment, I fixed the qindex to 200 in key frame.
Change-Id: I021f9ce4616ed3574c81e48d968662994b56a396
Replace error_resilient flag with use_prev_frame_mvs in
vp9_pick_inter_mode reference motion vector search selection.
This effectively turns off the simplified ref mv search in the
settings of frame resizing, even if error-resilient mode is off.
Change-Id: I7fed814ee7bc0cb419a03b846e0fc2de46ba7686
Update the frame motion vector only if previous frame motion vector
is needed for next frame reference motion vector.
Change-Id: Ica50f9d7b46ad4f815bba0d9e30f5546df29546f
The warning only happens in VP9 encoder's first pass due to src_mi
is not set up yet. But it will not fail the encoder as left_mi and
above_mi are not used in the first_pass and they will be set up again
in the second pass.
Change-Id: I12dffcd5fb1002b2b2dabb083c8726650e4b5f08
This allows us to track decode speed for new encodes so that we catch
problems like an encode change that makes decode really slow.
Change-Id: I92251a8b1f710b241f66e1042413df1b71b76038
This commit enables the use of sub8x8 blocks in RTC key frame
encoding. It requires the block size to be preset and will decide
the coding mode and encode the bit-stream.
Change-Id: I35aaf8ee2d4d6085432410c7963f339f85a2c19b
Rename set_modeinfo_offsets as set_mode_info_offsets, to be more
consistent with naming convention.
Change-Id: I68ca1f36c4a78127d9439a50c1506a2afd07927d
The later encoding process will take the top-left block's
mode_info for pre-determined block size.
Change-Id: I76a90f9ce7f3b2dbc2975b52442114e461c465b5
The restructure moves the decision into the rd pick
modes loop and makes a decision based at the 16x16
block level instead of only the 64x64 level.
This gives finer granularity and better visual results
on the clips I have tested. Metrics results are worse
than the old AQ2 especially for PSNR and this mode
now falls between AQ0 and AQ1 in terms of visual
impact and metrics results.
Further tuning of this to follow.
It should be noted that if there are multiple iterations
of the recode loop the segment for a MB could change
in each loop if the previous loop causes a change in the
complexity / variance bin of the block. Also where a block
gets a delta Q this will alter the rd multiplier for this block
in subsequent recode iterations and frames where the
segmentation is applied.
Change-Id: I20256c125daa14734c16f7cc9aefab656ab808f7
The function vp9_filter_block1d16_h8_ssse3 uses the PSHUFB instruction which has a 3 cycle latency and slows execution when done in blocks of 5 or more on Atom processors.
By replacing the PSHUFB instructions with other more efficient single cycle instructions (PUNPCKLBW + PUNPCHBW + PALIGNR) performance can be improved.
In the original code, the PSHUBF uses every byte and is consecutively copied.
This is done more efficiently by PUNPCKLBW and PUNPCHBW, using PALIGNR to concatenate the intermediate result and then shift right the next consecutive 16 bytes for the final result.
For example:
filter = 0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,8
Reg = 0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15
REG1 = PUNPCKLBW Reg, Reg = 0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7
REG2 = PUNPCHBW Reg, Reg = 8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15
PALIGNR REG2, REG1, 1 = 0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,8
This optimization improved the function performance by 23% and produced a 3% user level gain on 1080p content on Atom processors.
There was no observed performance impact on Core processors (expected).
Change-Id: I3cec701158993d95ed23ff04516942b5a4a461c0
Adds decoder config as a changeable parameter to unit tests, and
changes end to end test to use commonly used parameters to enable
base test of tiles encoding and frame parallel decoding.
Change-Id: I5d23a6857303b4d68b92b15c3f2f04a1bcb4c2bb
the flag in the header wasn't being set based on the encoder
configuration in non-intra only mode
broken since:
fbc2fbf Adding oxcf temp variable.
Change-Id: Ib4cff9901889824bc4e68d7f0f6deb1e41df2f53
The initial reset of this_rdc in vp9_pick_inter_mode is not needed,
since it will be re-assign when used.
Change-Id: Ic0e12d741cbab292fc214c1eabb48b129af7839b
Compare the current best mode rate-distortion cost with the skip
threshold to decide if performing motion search.
Change-Id: Ia071824f8dd3b7db485f424692a485a2da6a1a9f
These speed-up features for key frame coding are only turned on
in the settings of hybrid non-RD and RD mode decision. It provides
about 20% speed-up to the hybrid key frame coding at the expense
of certain compression performance loss. For vidyo1, the key frame
coding statistics are changed
9838F, 35.020 dB, 61677 us -> 9920F, 34.834 dB, 47556 us
Overall rtc set compression performance is down by -0.257%.
Change-Id: I0025447fda26bb7855e982955642b5f55d71b51f
When block size is below 16x16, the encoder swap from non-RD to
RD mode for key frame coding. This largely brough back the key
frame compression performance. For vidyo1 at 1000 kbps, the key
frame coding statistics are changed
9978F, 34.183 dB, 36807 us -> 9838F, 35.020 dB, 61677 us
As compared to the full RD case
7187F, 34.930 dB, 214470 us
The overall rtc set coding performance (single key frame setting)
is improved by 1.5%.
Change-Id: I78a4ecf025d7b24ec911e85be94e01da05e77878
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
No more checking of corrupted reference frame as we skip
decoding any non-intra frame in case of frame corrupted.
Change-Id: I77d41bbb02fc5f61972740e2d411441eb6a17073
Currently, VP9 supports column-tile encoding, which allows a frame
to be encoded in multiple column tiles independently. The number of
column tiles are set by encoder option "--tile-columns". This
provides a way to encode a frame in parallel.
Based on previous set of patches, this patch implemented the tile-
based multi-threaded encoder. Each thread processes one or more
tiles.
Usage:
For HD clips:
--tile-columns=2 --threads=1/2/3/4
While using 4 threads, tests showed that the encoder achieved
2.3X - 2.5X speedup at good-quality speed 3, and 2X speedup at
realtime speed 5.
Change-Id: Ied987f8f2618b1283a8643ad255e88341733c9d4
Change 71789 renamed CONFIG_VP9_HIGH to CONFIG_VP9_HIGHBITDEPTH.
However, one use of CONFIG_VP9_HIGH was missed.
Change-Id: I0ebb9c71380c6d810a25708d15471abf9533e695
the gtest implementation used only returns values between 0 and 2^31-1
+ temporarily disable some tests in fdct8x8_test which misbehave with the
new range
Change-Id: I45381076f0bea3317cc6728305890e4fd2f2facd
Currently, the configure script checks for x32 by testing just the
__ILP32__ define. However, on "plain" i386, __ILP32__ can also be
defined, for example by clang 3.5.0 and higher. (That gcc does not
define it there, is another issue, but not for this tracker.)
Therefore, extend the check by also checking for __x86_64__, which will
also be defined for x32.
BUG=887
Change-Id: I90ac1d6843caff0416e1dd360c0be3dbaa85c2ae
the gtest implementation used only returns values between 0 and 2^31-1
+ temporarily disable some tests in lpf_8_test which misbehave with the
new range
Change-Id: I8a026680c4b8c12dc14d4f24c33edb2315963114
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
This commit reworks the ONE_LOOP_REDUCED coefficient probability
model update process. It allows model update for every coefficient
across the spectrum at a coarser resolution, instead of performing
precise update only for certain subset of probability models.
The overall runtime remains nearly same (<1% change) for speed -6.
The compression performance is improved by 7.5% in PSNR for speed
-5 and 4.57% for speed -6, respectively.
Change-Id: Ifb17136382ee7e39a9f34ff4a4f09a753125c8d1
Synchronize all threads immediately as a subsequent decode call may
cause a resize invalidating some allocations.
fixes one aspect of crbug.com/437655
Change-Id: Ie993b62c2756478543206ddbe43ec6268d90a470
Change 72056 unfolded some macro definitions,
but lost some alternative behaviour required for
high bitdepth encodes.
This causes the encoder to crash, see issue 884.
Change-Id: I8ce4d73c9fe0a3c10ccb86fba210fabc8b2f0ccc
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 makes the codec automatically turn on error resilient
mode when using real-time mode for temporal scalable coding. It
fixes an enc/dec mismatch issue and re-enables the corresponding
unit test.
Change-Id: Ie1f7134e9a78ddd43e9b1555b3ee991c8a3afd0d
AQ2 modified to use mb_av_energy in defining variance
thresholds used alongside complexity when defining the
segment to be used for an SB64.
Slight improvements in metrics (ssim and PSNR).
Change-Id: Idb9cb73f7d9c4f7118cd7e84ac77b0f25cacbf81
Incorporate segment delta-q into estimated bits.
This generally improves the rate control under cyclic refresh (aq=3) mode.
Change-Id: I1dc60fb230e7d08357fae18909d8ed27bf58e037
A hidden enc/dec mismatch bug was accidentally triggered by
https://gerrit.chromium.org/gerrit/#/c/72247/
Adaptively adjust mode test kick-off thresholds in RTC coding
This commit temporarily turns off the broken unit tests to avoid
blocking other CLs while fixing.
Change-Id: I0a0f195030321190ce10879cd833187680576367
Probably not even the dominant platform the library is being built for.
Add --cpu= option description to help. The option already exists.
Don't allow passing just --cpu as a no-op.
BUG=826
Change-Id: Iaa3f4f693ec78b18927b159b480daafeba0549c0
This patch greatly increase the strength of AQ1.
Visual tests show strong gains on many clips but their is a big
hit on psnr.
SSIM is more mixed with some winners and losers.
Change-Id: Idaa5d3b41d8576096bfa000b62bc531c3d8bf6a1
Each tile's tok starting address is calculated before the encoding
process. These addresses are stored so that the same calculation
won't be done again in packing bit stream.
Change-Id: I0a3be0301f002260c19a850303f2f73ebc47aa50
When the golden frame is boosted, the rate correction factor is not
correlated well with other inter frames even in CBR mode. This commit
changes to use GF specific rate_correction_factor when gf_cbr_boost
is greater than 20%.
Change-Id: I6312c1564387bcacc11f4c5e8a9cfdc781b5c3ab
This commit allows the encoder to increase the mode test kick-off
thresholds if the previous best mode renders all zero quantized
coefficients, thereby saving motion search runs when possible.
The compression performance of speed -5 and -6 is down by -0.446%
and 0.591%, respectively. The runtime of speed -6 is improved by
10% for many test clips.
vidyo1, 1000 kbps
16578 b/f, 40.316 dB, 7873 ms -> 16575 b/f, 40.262 dB, 7126 ms
nik720p, 1000 kbps
33311 b/f, 38.651 dB, 7263 ms -> 33304 b/f, 38.629 dB, 6865 ms
dark720p, 1000 kbps
33331 b/f, 39.718 dB, 13596 ms -> 33324 b/f, 39.651 dB, 12000 ms
mmoving, 1000 kbps
33263 b/f, 40.983 dB, 7566 ms -> 33259 b/f, 40.978 dB, 7531 ms
Change-Id: I7591617ff113e91125ec32c9b853e257fbc41d90
This patch modified struct VP9_COMP. Created a struct ThreadData
to include data that need to be copied for each thread. In
multiple thread case, one thread processes one tile. all threads
share one copy of VP9_COMP,
(refer to VP9_COMP *cpi in the code)
but each thread has its own copy of ThreadData,
(refer to ThreadData *td in the code).
Therefore, within the scope of encode_tiles(), both cpi and td
need to be passed as function parameters.
In single thread case, the FRAME_COUNTS pointer in ThreadData
points to "counts" in VP9_COMMON.
Change-Id: Ib37908b2d8e2c0f4f9c18f38017df5ce60e8b13e
The intra mode penalty is covered by intra_cost_penalty. This
commit removes the other intra cost threshold, provided that the
constant 50 is negligible in normal rate-distortion cost.
Change-Id: I9b8b7483c43b9a41741622e7057def1f7d51bb72
This change is made in preparation for a
subsequent patch which adds acceleration
for the highbitdepth transform functions.
The highbitdepth transform functions attempt
to use 16/32bit sse instructions where possible,
but fallback to using the C implementations if
potential overflow is detected. For this reason
the dct routines are made global so they can be
called from the acceleration functions in the
subsequent patch.
Change-Id: Ia921f191bf6936ccba4f13e8461624b120c1f665
(cherry picked from commit 454342d4e7)
This commit makes a non-RD coding mode decision process for key
frame coding. It can be optionally turned on in speed -6 and above.
Change-Id: I0847258b392877a0210b4768bef88ebc9ad009b5
previously 'bit_depth_', which is later used to calculate 'mask_', would
be left uninitialized in non-high-bitdepth builds
Change-Id: Ia72035f4645baf3bb0f191504f491b934cdf1e0e
This commit allows more aggressive decision to skip forward
transform and quantization for luma component in RTC coding mode.
The chroma components remains going through the normal coding
routine, since they are not included in the non-RD mode search
process.
It reduces the runtime cost by 2% - 10%. In speed -6,
vidyo1 1000 kbps
16576 b/f, 40.281 dB, 8402 ms -> 16576 b/f, 40.323 dB, 7764 ms
nik720p 1000 kbps
33337 b/f, 38.622 dB, 7473 ms -> 33299 b/f, 38.660 dB, 7314 ms
dark720p 1000 kbps
33330 b/f, 39.785 dB, 13505 ms -> 33325 b/f, 39.714 dB, 13105 ms
The compression performance of speed -6 is improved by 0.44% in
PSNR and 1.31% in SSIM.
Change-Id: Iae9e3738de6255babea734e5897f29118bebc6d7
In AQ1 a rate adjustment was applied for blocks coded with a
deltaq. This tends to skew the partition selection and cause
rate overshoot.
For example, consider a 64x64 super block where some but not all
sub blocks are in a low q segment and some are in a high q segment.
The choice of Q when considering large partition and transform sizes
is defined by the lowest sub block segment id (currently this implies the
lowest Q). If some parts of the larger partition are very hard this will
cause a high rate component.
The correct behavior here is for the rd code to discard the large partition
choice and break down to sub blocks where some have low and some
have high Q. However the rate correction factor above mask the high
cost of coding at a larger partition size.
Change-Id: Ie077edd0b1b43c094898f481df772ea280b35960
Make the midpoint variance used in AQ mode 1 segmentation
depend on the overall complexity of the frame in two pass.
Change-Id: I452814ec57f7a32352e41bb250e78066abe952dd
By using 0xff for a short it was not setting the high bits. When
comparing the output with vtst to find non-zero elements it was skipping
vaules which had no low bits set such as -512 / 0xFE00.
Using -8191 as the first element of coeff will generate this condition.
BUG=883
Change-Id: Ia1e10fb809d1e7866f28c56769fe703e6231a657
All the assembly code has been removed, the tests no longer check for
the target, and android and chrome do not use the targets.
Change-Id: I193993f7b2b0bd6478453402f573ce3606e04e8d
Add an additional restriction to bit/complexity based
segmentation based on spatial variance.
Only lower Q when both the number of bits spent
in the initial encoding pass and the spatial complexity are
below a threshold. This will prevent the low Q segments
being used just because there is a surfeit of bits.
Small metrics gains especially opsnr.
derf ~0.2% std-hd ~0.3%
Change-Id: I6a8496d466d673f9b0e2b2ca6304ea7b6d8e1cce
This is the first of a series of patches to restructure and
improve AQ mode 1 (variance based AQ).
Change-Id: Idcf693131a3ea2459dcfd957a54a65b971fa4a2a
this was incorrectly set in test.mk by
93ffd37 Enable and fix resize_test for VP9
the test is now available when using --disable-vp9
Change-Id: I6acf44b0de647b34812ef5e18fd96447cdf9b25d
worker hooks return false on error, fix the assignment in Execute() used
in the TestSerialInterface test
Change-Id: I93c2e45f270330ae6d35a3a303411c4ee0f31337
Similar to mask_filter, the filter_cache in RD_OPT struct can be
moved out, and declared as a local variable since it is only
used in pick_inter_mode functions.
Change-Id: I412b99cca82bade07ac912064ec03dd1de6b2c17
Correct calculation of number of mbs in two pass code when
frame resizing is enabled. Always use initial number of mbs if
scaling is enabled, as this is what was used in the first pass.
Change-Id: I49a4280ab5a8b1000efcc157a449a081cbb6d410
The mask_filter in RD_OPT struct is used to record rd result in
filter decision. It is only used in pick_inter_mode functions,
and is removed from the struct and declared as a local variable.
Change-Id: I3c95c8632ba7241591ce00ef2ef5677b5e297d7b
The max_partition_size and max_partition_size are set at the
beginning while setting speed features, and then adjusted at
SB level. Moving them to mb struct ensures there is a local
copy for each thread.
Change-Id: I7dd08dc918d9f772fcd718bbd6533e0787720ad4
VP9/DatarateTestVP9Large.ChangingDropFrameThresh/[34] fails post the
merge of commit#ffa06b37. This commit adds reset of rc tracking info
when frame is dropped, and fixes the causes of the bad interaction
between the tests and the previous commit.
Change-Id: I848acfd9fcb336359662274325190f94aac76eae
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
In rare cases, the interaction between rate correction factor and Q
choices may cause severe oscillating frame sizes that are way off
target bandwidth. This commit adds tracking of rate control results
for last two frames, and use the information to prevent oscillating
Q choices.
Change-Id: I9a6d125a15652b9bcac0e1fec6d7a1aedc4ed97e
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
Current setting had active_worst_quality set too high (close to worst_quality)
for first frame(s) following first key frame. This changes that to be somewhat
more aggressive in allowing active_worst_quality to be lower following key frame.
Also remove the 4/5 reduction in active_worst for key frame as
this should be set by the user qp_max setting.
Change-Id: I0530b3ddcc85c00e3eb7568de1b14a31206c4a4c
The function pointer in compressor instance does not change, so this
commit changes to call the function directly.
Change-Id: I9c9c460e3475711c384b74c9842f0b4f3d037cc5
This commit adds a check condition to the prediction buffering
operation used in the rtc coding mode. This resolves a unit test
warning in example/vpx_tsvc_encoder_vp9_mode_7.
Change-Id: I9fd50d5956948b73b53bd8fc5a16ee66aff61995
These 2 members in RD_OPT were moved to TileDataEnc struct
already, and therefore were removed here.
Change-Id: I22fee3b67f96e473a58e194a7edc76dbd48bfa04
Several frame counters in encoder are updated at SB level. Combine
those counters and put them in a separate struct, which allows us
to allocate one copy for each thread.
Change-Id: I00366296a13c0ada4d8fa12f5e07728388b6cab7
Modified VP9_COMP struct to include MACROBLOCK *mb. This change
makes it feasible in multi-thread case to allocate a mb for each
thread.
Change-Id: I624d6d1aa9c132362200753e5d90b581b1738d6e
A flush bug is discovered during putting frame parallel decoder
into Android. This test will expose that bug.
Change-Id: Ia047f27972f4da0471649f79f1f91e7695297473
Two members in struct CYCLIC_REFRESH
int64_t projected_rate_sb;
int64_t projected_dist_sb;
are updated at the superblock level, which makes them shared data
in the multi-thread situation, and requires extra work to handle
them. However, those values are updated and used immediately, and
therefore can be removed. This patch cleaned up the code and
removed the two members.
Change-Id: I2c6ee4552bf49fb63ce590cdb47f9723974fffb1
Prepare for the introduction of frame-size change
logic into the recode loop.
Separated the speed dependent features into
separate static and dynamic parts, the latter being
those features that are dependent on the frame size.
Change-Id: Ia693e28c5cf069a1a7bf12e49ecf83e440e1d313
Add extra vp9_clear_system_state() calls to fix
double / mmx issue introduced into first pass
code for 32 bit builds.
Change-Id: I84cd2986b80d83650a091ab25c43755efeb82e03
This reverts commit 7d07f512cd.
this breaks visual studio builds:
'#' : invalid character : possibly the result of a macro expansion
Change-Id: I77170d549afb71e75a878fa0f6acd204fe8d9e67
Rate correction factor is used to correct the estimated rate for any
given quantizer, and feeds into rate control for quantizer selection.
We make use of the actual bits used to calculate this rate correction
factor with an adjustment limit to prevent over-adjustment.
This commit adapts the adjustment limit to the difference between the
estimated bits and the actual bits, allows the adjustment limit to vary
between 0.125 (when estimate is close to actual) and 0.625 (when there
is >10X factor off between estimated and actual bits). By doing this,
the commit appears to have largely corrected two observed issues:
1. Adjustment is too slow when the actual bits used is way off from
estimate due to the small adjustment limit.
2. Extreme oscillating quantizer choices due to the feedback loop.
Change-Id: I4ee148d2c9d26d173b6c48011313ddb07ce2d7d6
This commit makes the speed -6 and above use the reconstructed
boundary pixels for precise intra prediction. This allows more
intra prediction modes to be tested in the non-RD coding process.
Enabling horizontal and vertical intra prediction modes can
improve the speed -6 compression performance for rtc set
by 0.331%.
Change-Id: I3a99f9d12c6af54de2bdbf28c76eab8e0905f744
I0c5f010 changed to allow update golden reference buffer in CBR mode,
this commit changes the use of rate_correction_factor for those frames
to be aligned with the new usage. This commit attempts to solve two
issues:
a. Initialization of rate correction factor for Golden Frame
Prior to this patch, even the regular inter frame has been update
the rate correction factor based on content and encoding results,
the first golden frame would still use the ininitialized value
that can be way off.
b. Allowing rate correction factor update to be slightly faster
Prior to this patch, when the rate correction factor is off, the
update to the factor is too slow, the factor could not get close
to a semi-correct value even after many frames.
The commit helps all clips in psnr/ssim metric, but especially to
a few clip in RTC set that rate correction was way off. For example
thaloundeskmtgvga gained about .5dB for both overall/average psnr.
Change-Id: I0be5c41691be57891d824505348b64be87fa3545
Adds support for one-pass rc-enabled SVC encoder with callbacks for
getting per-layer packets.
- the callback function registration is implemented as an encoder
control function.
- if the callback function is not registered, the old way of
aggregating packets with superframe will take effect.
- one more control function “VP9E_GET_SVC_LAYER_ID” has been
implemented to get the temporal/spatial id from the encoder
within the callback. This can be used to get the ids to put on RTP
packet.
Change-Id: I1a90e00135dde65da128b758e6c00b57299a111a
This commit rename a reserved color space entry to BT_2020, it intends
to provide support for VP9 bitstream to pass along the color space
type defined in BT.2020(Rec.2020)
please note this entry does not have any effect on encoding/decoding
behavior, but allow applications to the pass the information along
from encoding end to decoding end.
Change-Id: I4678520e89141ea5e8900f7bd1c0e95b710b7091
This commit integrates the non-RD mode decision process and the
encoding process into a single recursion scheme.
Change-Id: I6a7e72a0b84d567554801ebbe01ec75d54c1f77d
The obj_int_extract code is no longer worth maintaining. It creates
significant issues when adapting for different build systems and no
longer offers as significant of a performance benefit due to
improvements in intrinsics.
Source files will remain until the various third-party builds are updated.
The neon fast quantizer has been moved to intrinsics. The armv6 version
has been removed because so few remaining targets require it.
Compilers and processors have improved significantly since the
pack_tokens code was written. The assembly is no longer faster than the
C code.
pack_tokens were the only optimizations for the armv5te targets so the targets
will be removed after the test infrastructure has been updated.
BUG=710
Change-Id: Ic785b167cd9f95eeff31c7c76b7b736c07fb30eb
This patch was to fix the vpxdec fuzzing3 test failure. When an
error occurs, setjmp() is invoked, which calls the decoder
removing routine. In multiple thread situation, other threads
could try to access the frame context memory that is already
deallocated, thus causing a segfault.
An invalid unit test was added for this issue.
Change-Id: Ida7442154f3d89759483f0f4fe0324041fffb952
The aim of this patch is to apply a positive weighting to
frames that have a significant number of blocks that are
of low spatial complexity and are dark. The rationale behind
this is that artifacts tend to be more visible in such frames.
In this patch the weight is only applied in regard to the distribution
of bits between frames. Hence if all the frames share similar
characteristics (as is the case for most of our short test clips) there
will be little or no net effect.
However, the effect can be seen on some longer form test content.
For example Tears of steel baseline test:
2323.09 Kbit/s opsnr 39.915 ssim 74.729
With this patch:-
2213.34 Kbit/s opsnr 39.963 ssim 74.808
(Sligtly better metrics and about 5% smaller)
The weighting may well need some further tuning along side changes
to the aq modes.
Change-Id: Ieced379bca03938166ab87b2b97f55d94948904c
This commit removes the cyclic aq mode dependency on
in_static_area and reworks the corresponding cut-off thresholds.
It improves the compression performance of speed -5 by 1.47% in
PSNR and 2.07% in SSIM, and the compression performance of speed
-6 by 3.10% in PSNR and 5.25% in SSIM. Speed wise, about 1% faster
in both settings at high bit-rates.
Change-Id: I1ffc775afdc047964448d9dff5751491ba4ff4a9
the offending assembly code was deleted in:
08e38f0 VP8 for ARMv8 by using NEON intrinsics 14
the intrinsics currently pass.
fixes issue #725
Change-Id: I43e4263bef21f9d9008c51ffdfa39fcf10b8e776
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
The test filter is not a prefix matcher. It requires test type to
contain no more than the optimization type. In this example, SSSE3_64
fails to match and the test is not skipped even when SSSE3 is not
available.
Change-Id: Ia74229a167c88da4e6da169012a7a77d438c3f75
Check that the numerator is not zero. If it is, guess 30fps.
Fixes a clang IOC error in the quantize test. It's very unlikely for
this to occur in the wild because the setup in the quantize test is very
nonstandard.
Change-Id: Icdab7b81d4e168d3423e14db20787f960052e0c3
This commit makes the RTC coding mode to conditionally skip the
reference frame mode search, when the predicted motion vector of
the current reference frame gives more than two times sum of
absolute difference compared to that of other reference frames.
It reduces the runtim by 1% - 4% for speed -5 and -6. The average
compression performance is improved by about 0.1% in both settings.
It is of particular benefit to light change scenarios. The
compression performance of test clip mmmovingvga.y4m is improved by
6.39% and 15.69% at high bit rates for speed -5 and -6, respectively.
Speed -5
vidyo1 16555 b/f, 40.818 dB, 12422 ms ->
16552 b/f, 40.804 dB, 12100 ms
nik 33211 b/f, 39.138 dB, 11341 ms ->
33228 b/f, 39.139 dB, 11023 ms
mmmoving 33263 b/f, 40.935 dB, 13508 ms ->
33256 b/f, 41.068 dB, 12861 ms
Speed -6
vidyo1 16541 b/f, 40.227 dB, 8437 ms ->
16540 b/f, 40.220 dB, 8216 ms
nik 33272 b/f, 38.399 dB, 7610 ms ->
33267 b/f, 38.414 dB, 7490 ms
mmmoving 33255 b/f, 40.555 dB, 7523 ms ->
33257 b/f, 40.975 dB, 7493 ms
Change-Id: Id2aef76ef74a3cba5e9a82a83b792144948c6a91
This commit unfolds the legacy macro definitions used in the
sub-pixel motion search and refactors the operational flow for
later optimizations.
Change-Id: I3e3f770cad961d03d1a6eb0b2a0186cc77eaf2b8
The current logic was allowing for disabling golden refresh only
for two pass svc encoding. This change disables it as long as
more than 1 layer encoding is used (for example temporal layers under 1pass CBR).
Change-Id: I4dc5204a7ad365c821ec7963e93b59da82e1826b
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
A recent change has introduced big quality drops for speed 7 and 12
for --rt mode. The change reverted the big drop and improved quality
by 9.5% for speed 7 and 13.4% for speed 12.
Change-Id: I07b82e3bb6002a73af486a083458c88877bdad01
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
Use intrinsics for neon quantization. Slight loss (<5%) of performance
compared to the assembly. Roughly 10x faster on arm64 because that was
running C code before.
Change-Id: I7cf5242d8f29b7eab5bca6a1c20c89c9fc9ca66d
This commit makes the inter prediction buffer system to support
hybrid partition search. It reduces the runtime of speed -5 by
about 3%. No compression performance change.
vidyo1 720p 1000 kbps
11831 ms -> 11497 ms
nik 720p 1000 kbps
10919 ms -> 10645 ms
Change-Id: I5b2da747c6395c253cd074d3907f5402e1840c36
Combined vp9_denoiser_8xM_sse2 and vp9_denoiser_4xM_sse2 into one
function vp9_denoiser_NxM_sse2_small and passed the bitexact testing.
Changed the name of the function vp9_denoiser_64_32_16xM_sse2 to
vp9_denoiser_NxM_sse2_big.
Change-Id: Ib22478df585994dd347ebae04202c0b701e7f451
This commit changes to allow the usage of golden reference frame in
VP9 CBR mode to improve quality. VP9 supports potentially up to 8
reference buffers, it has reference buffers available for this
purpose. This was not possible in VP8 as golden and alt-ref buffers
were used for temporal scalability purpose in CBR mode in WebRTC.
For frames that update golden frame, there can be a quality boost.
The amount of allowed bitrate boost can be controlled via parameter
rc_max_inter_bitrate_pct. The inital value of the boost ratior is
currently based on over_shoot_pct. Further experiments will work
out the adaption of this boost value.
Change-Id: I0c5f010c8fd8b7b598f69779c1b30e5b2ac30a4d
Added code to relax the active maximum Q in response
to extreme local overshoot to reduce bandwidth peaks.
The impact is small in metrics terms, but it this helps reduce
bandwidth spikes and overall overshoot in a number of
clips in our tests sets (especially the YT test set).
In particular this should help prevent very big spikes where a clip
is mainly easy but has a short hard section. In such a case a choice
of maximum Q for the clip as a whole may allow us to hit the overall
target rate but give some extreme spikes. The chunked encoding in YT
mitigates this problem but it can show up where a longer clip is
coded as a single chunk.
Change-Id: I213d09950ccb8489d10adf00fda1e53235b39203
The zero motion vector was effectively used in the subsampled pixel
based variance calculation. This commit makes it directly use zero
mv to generate prediction.
Change-Id: Ica83dc843e9f8da2f89c3ef451e50f16214c0def
0 means that golden boost is off, and uses average frame target rate,
a non-zero number means the percentage of boost over average frame
bitrate is given initially to golden frames in CBR mode.
Change-Id: If4334fe2cc424b65ae0cce27f71b5561bf1e577d
-Use full bandwidth (when temporal layers is on) for checking switching.
-Normalize metric wrt num_blocks.
-Rounding fix to update of average noise level metric.
-Make default internal denoiser mode == kDenoiserOnYUV (in denoiser set_parameters()).
-Adjust some thresholds.
Change-Id: Ib827512b25a7bf1f66c76d3045f3a68ce56b1cd2
The point at which frames are scaled to their
coded dimensions is moved into the re-code loop.
This is in preparation for a further patch that
will add logic into the re-code loop to reduce
the coded frame size if the encoder is struggling
to hit the target data rate at the native frame
size.
Change-Id: Ie4131f5ec6fb93148879f6ce96123296442bf2d1
Add second level arf Q adjustment when using dual arfs
in constant Q mode.
Previously in constant Q mode enabling dual arf hurt by ~5%
but with this change the average benefit is ~1-1.5% with some
mid range data points up ~10%.
Note however that it still hurts on some clips including
some very low motion show content.
Change-Id: I5b7789a2f42a6127d9e801cc010c20a7113bdd9b
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
The initialization of this_mode_pred does not work when the ref_frame
loop ever goes beyond LAST_FRAME. This commit fixes the subtle issue
and allows potentially expanding the loop to test GOLDEN_FRAME.
Change-Id: Ibbd427a22160d1d9eacb8ed0c87f88d6cef9c0f3
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
This commit refactors the rate distortion structure used in the
non-RD coding mode and saves a few RDCOST calculations.
Change-Id: I62c3416c300d2c5372f21b96d93a6b633a34ab3a
The existing speed features produce horrible encoding results, almost
30% worse than cpu-used=4, this commit adjust the speed features to
produce relatively resonable results to be within 3%-5% of cpu-used=4.
Change-Id: I0ca6ebafb33024d4a0cbcf04c78a4a00b8dd1ecf
Its functionality has been replaced with choose_partitioning and
threshold based control on split mode check.
Change-Id: Ic9bb321df06b524f5c38ea5874dc6f6a8f93c5e3
This speed feature has been deprecated in both yt and rtc coding
modes. This commit removes the related operations.
Change-Id: I079c79c6adafe45581af2ebf8b98faebcface1ce
This commit re-designs the recursive partition search scheme in
rtc speed -5. It first checks if the current block is under cyclic
refresh mode. If so, apply recursive partition search. Otherwise,
perform sub-sampled pixel based partition selection. When the
pre-selection finds the partition size should be 32x32 or above,
use the partition size directly. Otherwise, apply partition search
at nearby levels around the preset partition size.
It is enabled in speed -5. The compression performance of rtc
speed -5 is improved by 9.4%. Speed wise, the run-time goes slower
from 1% to 10%.
nik_720p, 1000 kbps
33220 b/f, 38.977 dB, 10109 ms -> 33200 b/f, 39.119 dB, 10210 ms
vidyo1_720p, 1000 kbps
16536 b/f, 40.495 dB, 10119 ms -> 16536 b/f, 40.827 dB, 11287 ms
Change-Id: I65adba352e3adc03bae50854ddaea1b421653c6c
Extend --auto-alt-ref from parameter so we can use it to
turn multi-arf on and off from the command line.
For now the range is 0-off, 1-on, 2-multi-arf on.
Rename play_alternate to enable_auto_arf
Change-Id: Id7b64407cfbe76ba0090a83b588a03e22a240386
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
Currently, the tokens for a tile are stored immediately after its
preceding tile, which causes a dependency. This is unnecessary
since we always allocate enough memory for tokens. Removing
the dependency allows token writing done in parallel. This patch
doesn't change encoding result.
Change-Id: I7365a6e5e2c2833eb14377c37e1503c9d0f26543
This should be set right after decoder really start to decode frame
instead setting at the end.
Even decoder does not have a displayable frame to show and return NULL
to application, this should be set too.
Change-Id: If0313a834bc64e3b0f05a84f4459d444d9eab0d8
When early termination is triggered, properly reset the rate cost
to invalid value to avoid potential ioc issue.
Change-Id: I3444390be2e49a34bb02cf8a74c33d5dbd96d88d
Delete gfboost_qadjust() and move Q based adjustment
into calc_frame_boost(). Also remove clamping. Making
the adjustment here means that it influences not just the
boost level but also the selection of the GF/ARF interval.
This change gives a small average gain in PSNR but
larger gains in SSIM, especially for harder std-hd set (1.5%)
Change-Id: I3aa81b8feccaeff93d915e19fb9cf5cd64c86327
Covers all profiles and input formats. The tests check if the
encode succeeds and if the psnr is sane.
Change-Id: I195a5330debf92562846121819b6eaf961e27c01
This commit fixes an ioc issue that will happen when the cumulative
variables are not in effective use. The fix discards these
redundant additions.
Change-Id: Idbac5bfb989c0cedc5f8a323effce938519b2457
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
move them from VP9Worker::data[12] to allow the structure to be reused a
bit more naturally by the multi-threaded loopfilter.
Change-Id: I31b49c9e93ca744fd7f6d6ed8696671188fb2c1d
This removes an unnecessary restriction that causes
a problem (noticed by AWG) when the forced key frame
interval is set to a very small value, such as 10. In this case
we were being forced to code minimal length GF groups.
Change-Id: I76ef5861a09638ff51f61fea02359554184ada53
We encode a empty invisible frame in front of the base layer frame to
avoid using prev_mi. Since there's a restriction for reference frame
scaling factor, we have to make it smaller and smaller gradually until
its size is 16x16.
Change remerged.
Change-Id: I9efab38bba7da86e056fbe8f663e711c5df38449
This reverts commit 452dc21500.
This change has introduced a significant quality regression on content
with forced key frames. (e.g. the YT and yt-hd set). It is most
noticeable in static content where the kf bits dominate. Here, despite
key frames being apparently coded at the same Q, there is a drop in all
metrics of ~20% (e.g clXR and BFa0).
Change-Id: Iba14cc61778c0846fa0a59c33c55a9fc49512cb4
Compare the estimated rate and distortion to the thresholds scaled
according to the operating block size and determine if further
split partition search will be run. The compression performance of
speed -5 is changed by -0.074%. The encoding speed is 10% - 15%
faster.
vidyo1 720p
16545 b/f, 40.492 dB, 11475 ms -> 16535 b/f, 40.486 dB, 10100 ms
nik720p
16624 b/f, 36.310 dB, 10071 ms -> 16617 b/f, 36.313 dB, 8346 ms
Change-Id: Ic9197ab5761279ae55d2fb7813b2af0e0db497b8
Reduce the intra_cost_penalty for non-rd mode,
and some updates to VAR_BASED_PARTITION.
Visual tests show some improvement at Speed 6, for RTC clips.
Change-Id: If9090daf7aed14906a32d931a538ab544bbca606
This commit replaces the use of copy_partitioning with
choose_partitioning based on the sse of subsamped pixels, which
provides significantly better coding performance and runs at
similar speed, as compared to copy_partitioning. It improves rtc
speed 5 coding performance by 3%.
Change-Id: I52d3682a12dce0147f5e52383a594fc242ca3228
this change checks that CONFIG_SPATIAL_SVC is defined and adds a TODO to
ensure this is changed in the future as the release headers can't
depend on vpx_config.h.
vpx/vpx_encoder.h:164:5: warning: "CONFIG_SPATIAL_SVC" is not defined
[-Wundef]
Change-Id: I797a0150e5f56caf048e7ee00b282fbc9c5ede19
We encode a empty invisible frame in front of the base layer frame to
avoid using prev_mi. Since there's a restriction for reference frame
scaling factor, we have to make it smaller and smaller gradually until
its size is 16x16.
Change-Id: I60b680314e33a60b4093cafc296465ee18169c19
Move the point at which input frames are scaled
into the recode loop. This will allow us to change
the coded frame size dynamically in response
to previous attempts to encode the frame at a
higher resolution.
A following patch will implement a scheme for
resizing the frame in the recode loop.
Change-Id: I6a59c02d6ac1626512edad6de8b60063b79433e6
This commit makes a struct that contains rate value, distortion
value, and the rate-distortion cost. The goal is to provide a
better interface for rate-distortion related operation. It is
first used in rd_pick_partition and saves a few RDCOST calculations.
Change-Id: I1a6ab7b35282d3c80195af59b6810e577544691f
Add back clamp which ensures that the Q adaptation
is turned off when the over_shoot_pct and under_shoot_pct
parameters are set to 100.
Change-Id: Id0161b114d39a3029cd3eb28020caab0c3914922
There are two CreateDecoder functions and decode_test_driver is not
calling the right function now. This bug is discovered during really
enable the frame parallel flag inside libvpx. This bug does not affect
any existing unit test though.
Change-Id: Icd9633c4b66d50e422a09c4310ff791082878936
Make sure VP9 frame-parallel decode passes all the standard
test vectors. Only test running with 2,3,4 threads now.
Also refactor the video decode test driver to support passing
in decode flags which is used to enable frame-parallel decode.
Change-Id: I6a712464232c2e13681634951c7e176312522e1e
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
pthread.h is not supported in windows. vp9_thread.h includes
the emulation layer for pthread in windows.
Change-Id: I2b1c8ec299928472faca7ebeea998170c9f4d744
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
The relationship of the user private data at runtime
is not preserved from decode() to this call which may
occur at an unknown point in the future
Change-Id: Ia7eb25365c805147614574c3af87aedbe0305fc6
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
Some files were not shown because too many files have changed in this diff
Show More
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
