Commit Graph

5890 Commits

Author SHA1 Message Date
Marco
a304d26bf1 vp9-svc: Fix to resetting pattern/flags on key frame.
Chane only affects 1 pass cbr svc mode.

Change-Id: Ie28557409eb87673ed0b66c6dfe1bf3509a18c4d
2016-03-07 13:54:43 -08:00
Marco
2b5ab8095e vp9-svc: Non-rd pickmode, fix condition for testing zeromv for spatial ref.
Change-Id: I75bad321d5814a6df6735a8ead7700ab93b50763
2016-03-07 09:55:25 -08:00
Paul Wilkins
38b3593eb9 Merge "Fix bug in stats output for HBD." 2016-03-04 15:55:11 +00:00
paulwilkins
c7780075ec Fix bug in stats output for HBD.
Change-Id: Ic3b033e22cdcdc4a0f78cef433b2208fd254c27b
2016-03-04 12:57:10 +00:00
Marco
38e401b5da vp9-svc: Fix issues with svc with periodic key frames.
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
2016-03-02 16:59:58 -08:00
Alex Converse
fcaedc5681 Merge "Simplify the inner loop of duplicate_mode_info_in_sb." 2016-03-02 17:21:08 +00:00
Alex Converse
8b47c2662a Merge "Port "cost_coeff speed improvements" to vp9." 2016-03-02 17:20:58 +00:00
Alex Converse
6c0dd11f29 Simplify the inner loop of duplicate_mode_info_in_sb.
Change-Id: I8ba9164c3550379fb998d4d074846e90fb2c6f8c
2016-03-01 17:43:02 -08:00
Marco
c83bcb3474 vp9-svc: Allow for 2 stage downscaling for spatial layers.
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
2016-03-01 15:48:42 -08:00
Jingning Han
dbaf5f64e2 Merge "Port alt reference frame filter improvement to VP9" 2016-03-01 22:31:37 +00:00
Jingning Han
405fe6b78c Merge "Back port partition rate cost estimate fix to VP9" 2016-03-01 22:31:30 +00:00
Jingning Han
9c0a2a7935 Port alt reference frame filter improvement to VP9
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
2016-02-29 16:55:13 -08:00
Jingning Han
8d006a16c5 Back port partition rate cost estimate fix to VP9
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
2016-02-29 16:50:35 -08:00
Alex Converse
ee5d3d3760 Don't copy tx_size to all 8x8 MIs in the prediction block.
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
2016-02-29 15:23:21 -08:00
Marco
55a09f7f45 vp9-svc: For 1 pass svc, remove frame-level upsampling.
With the svc fix in https://chromium-review.googlesource.com/#/c/328978/,
the asan error is resolved, so this should work now.

Change-Id: I57b2a593651d414e1b445431d90f2fdc3281128b
2016-02-29 08:56:14 -08:00
James Zern
8062e10162 Revert "vp9-svc: Fix speed issue with source downscaling for spatial layers."
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
2016-02-26 16:53:26 -08:00
Julia Robson
74a679de6f Port "cost_coeff speed improvements" to vp9.
About a 5% faster overall encode (perf cycles) at speed zero!

Change-Id: Iaf013ba75884415cd824e98349f654ffb1c3ef33
2016-02-26 14:47:18 -08:00
Marco Paniconi
a69c3f2823 Merge "vp9-svc: Bugfix for svc in non-rd variance partition." 2016-02-26 22:39:28 +00:00
Marco
6a23966c34 vp9-svc: Bugfix for svc in non-rd variance partition.
Reset the scale factors before build_inter_predictors.

Add datarate tests for 3 spatial layers, which exposed this issue.

Change-Id: I7f81efbe44345ecea9fdd5f639a4cca76aed3874
2016-02-26 09:24:18 -08:00
Marco
f51f0998e1 vp9-svc: Fix speed issue with source downscaling for spatial layers.
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
2016-02-26 08:11:37 -08:00
James Zern
48755f9f1a Merge "vp9/10: fix forced keyframes w/alt-refs enabled" 2016-02-26 03:52:44 +00:00
James Zern
ac4c37c684 vp9/10: fix forced keyframes w/alt-refs enabled
in 1-pass encodes. issues with 2-pass as well as other forced flags
persist.

Change-Id: Ic7ceb906fccea6456d5df96483c10cacd46e01c7
2016-02-24 15:56:37 -08:00
Marco
1cab19e9e4 vp9: Update to scene/content change detection.
Update some parameters and put check on frame size change.
For 1 pass VBR mode, speed >= 5.

Change-Id: If24ed94a21e705ea57d40b9bf235ad079db786fc
2016-02-23 17:52:37 -08:00
Marco
a0278cad3f vp9 aq-mode=3: Allow it to be used for 1 pass VBR mode.
Change-Id: I630b8e33106c78382545d49da5fb4c75b1b0b528
2016-02-22 17:44:20 -08:00
Marco
958f372509 vp9: 1 pass vbr real-time mode: Adjust gf refresh for scene change.
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
2016-02-19 09:40:23 -08:00
James Zern
b6b40d3015 Merge changes from topic 'vp9-missing-alloc-checks'
* changes:
  vp9_svc_layercontext: add missing alloc checks
  vp9_resize: add missing alloc checks
  vp9_encoder: add missing alloc checks
  vp9/decoder,resize_mv_buffer: add missing alloc check
  vp9_cyclic_refresh_alloc: correct cleanup on error
2016-02-18 18:53:13 +00:00
James Zern
00a39d6e04 vp9_svc_layercontext: add missing alloc checks
Change-Id: Ieb305c178c64b3864f0e8fc1a13c29799e1ae8f5
2016-02-17 16:12:40 -08:00
James Zern
efbab73cc1 vp9_resize: add missing alloc checks
Change-Id: I87ef6dec7cd2e8f9a40135e5ca11b13520ebd6d7
2016-02-17 16:12:27 -08:00
Alex Converse
09f9c5d7f9 Better workaround for Bug 1089.
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
2016-02-17 14:46:26 -08:00
James Zern
1710f6507d vp9_encoder: add missing alloc checks
Change-Id: I08939ab22446b86fae884daaebd1f14c49b21897
2016-02-17 14:33:38 -08:00
James Zern
7997c68ed4 vp9_cyclic_refresh_alloc: correct cleanup on error
previously only the CYCLIC_REFRESH allocation was being freed

Change-Id: I6e1783d077c5ca83c8d62ea9642f1fb03f2e5bf3
2016-02-17 12:36:49 -08:00
James Zern
73f45fcf0b vp9,encoder: relocate setjmp
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
2016-02-16 19:25:54 -08:00
Marco
89cc682528 vp9-real-time mode: Fix condition for allowing reference masking.
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
2016-02-16 13:10:27 -08:00
James Zern
548889c35a Merge "Fix tsan error in VP9 sub8x8 intra mode search" 2016-02-16 20:26:42 +00:00
Jingning Han
d642294b1c 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
2016-02-16 11:24:09 -08:00
Marco Paniconi
0546636c74 Merge "vp9-resize: Fix an issue with external dynamic resize." 2016-02-16 16:18:43 +00:00
Marco
3cbc26f31b vp9-resize: Fix an issue with external dynamic resize.
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
2016-02-12 15:06:48 -08:00
James Zern
6c323e359a vp9_receive_raw_frame: add missing setjmp
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
2016-02-11 19:20:12 -08:00
Marco
34d12d1160 vp9-resize: Force reference masking off for external dynamic-resizing.
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
2016-02-11 08:35:57 -08:00
Alex Converse
0091fef7a2 Merge "Restore previous motion search bit-error scale." 2016-02-10 01:21:01 +00:00
Alex Converse
fac947df77 Restore previous motion search bit-error scale.
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
2016-02-09 13:20:25 -08:00
Marco
f288c943c4 vp9-dynamic resize: Fix bug on releasing scaled reference.
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
2016-02-09 11:10:28 -08:00
Marco
aff0a802e7 vp9-aq-mode=3: Reset cyclic refresh under config/external size change.
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
2016-02-09 10:10:36 -08:00
Yaowu Xu
5c327cb450 Set use_highbitdepth flag to either 1 or 0
This reduces the buffer allocation size for high bit depth buffers by
more than 70%.

Change-Id: Iab9007d9684e8ac30ce7fa94e61f038f01792fc5
2016-02-08 20:30:50 -08:00
Marco
2a3f994723 vp9-denoiser: Allow for checking skin for bsize <=32x32.
Change-Id: I6c99ac6fd3735e24cf668479c55bff313c0b4970
2016-02-08 13:01:38 -08:00
Jingning Han
f032c7eaed Merge "Account for sub8x8 block skip mode cost in RD decision" 2016-02-08 19:40:01 +00:00
Jingning Han
203bdd20fb Account for sub8x8 block skip mode cost in RD decision
Make this consistent with regular block size rate-distortion
optimization. It improves the compression performance:
derf    0.055%
hevcmr  0.129%

Change-Id: I112fe734f592c21bc7aa6efb7e3f269c4214ee7b
2016-02-08 10:18:51 -08:00
Jingning Han
ac6d40ece8 Clean up in vp9_rd_pick_inter_mode_sb
Use local variable.

Change-Id: I0d3df36cf4536958a0cda422f6c30da50f0e0bbf
2016-02-08 10:15:02 -08:00
Marco
ae98a61e22 vp9-dynamic resize: On a resize trigger frame refresh alt_ref.
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
2016-02-05 16:32:49 -08:00
Alex Converse
def519efb3 Fix an uninitialized warning.
Change-Id: I3ac1070df2948f6d8637ade7192400be53a7675c
2016-02-05 12:06:38 -08:00