]> granicus.if.org Git - libvpx/log
libvpx
8 years agovp9: Move the initialization of some rc variables to rc_init.
Marco [Fri, 29 Apr 2016 00:06:18 +0000 (17:06 -0700)]
vp9: Move the initialization of some rc variables to rc_init.

And initialize a variable (avg_source_sad) that was not initialized to 0.

Change-Id: I63223e20fffe2a2ea38f03da62dad16af0618ea4

8 years agoMerge "Avoid an unsigned overflow in invert_quant"
Alex Converse [Thu, 28 Apr 2016 19:44:03 +0000 (19:44 +0000)]
Merge "Avoid an unsigned overflow in invert_quant"

8 years agoMerge "vp9: Don't force block size to 8x8 in denoiser."
Jacky Chen [Thu, 28 Apr 2016 18:06:29 +0000 (18:06 +0000)]
Merge "vp9: Don't force block size to 8x8 in denoiser."

8 years agoMerge "Port convolve test refactor to master."
Alex Converse [Thu, 28 Apr 2016 16:26:53 +0000 (16:26 +0000)]
Merge "Port convolve test refactor to master."

8 years agoMerge "bitreader: remove an unsigned overflow."
Alex Converse [Thu, 28 Apr 2016 16:26:37 +0000 (16:26 +0000)]
Merge "bitreader: remove an unsigned overflow."

8 years agovp9: Don't force block size to 8x8 in denoiser.
JackyChen [Thu, 28 Apr 2016 05:18:11 +0000 (22:18 -0700)]
vp9: Don't force block size to 8x8 in denoiser.

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

8 years agoPort convolve test refactor to master.
Alex Converse [Wed, 27 Apr 2016 23:53:33 +0000 (16:53 -0700)]
Port convolve test refactor to master.

Brings f03e238f to master.

Change-Id: I7f7754e7d1288b103a4510303d10afc68a7d8ca8

8 years agoTweak casts on vpx_sub_pixel_variance to avoid implicit overflow.
Alex Converse [Thu, 19 Nov 2015 23:57:57 +0000 (15:57 -0800)]
Tweak casts on vpx_sub_pixel_variance to avoid implicit overflow.

Change-Id: I481eb271b082fa3497b0283f37d9b4d1f6de270c

8 years agobitreader: remove an unsigned overflow.
Alex Converse [Fri, 20 Nov 2015 00:09:03 +0000 (16:09 -0800)]
bitreader: remove an unsigned overflow.

bits_left is in the range [0, 64 (= BD_VALUE_SIZE)] , so the narrowing
conversion should be safe.

Change-Id: I943fcd359eaad76249ee1e1fb03a2ac16945d2fd

8 years agoBe explicit about overflow in vpx_variance16x16_sse2.
Alex Converse [Thu, 19 Nov 2015 23:43:24 +0000 (15:43 -0800)]
Be explicit about overflow in vpx_variance16x16_sse2.

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

8 years agoRemove casts on < 16x16 variance.
Alex Converse [Thu, 19 Nov 2015 23:39:40 +0000 (15:39 -0800)]
Remove casts on < 16x16 variance.

These blocks will never overflow since max sum is +/-255*w*h.

Change-Id: Ia2c630339fd9cfb411b56b6040ff402095f12a2e

8 years agoAvoid an unsigned overflow in invert_quant
Alex Converse [Fri, 20 Nov 2015 00:46:46 +0000 (16:46 -0800)]
Avoid an unsigned overflow in invert_quant

Change-Id: I16a570b2af66b6580d1cd6f8345a25f079009bf4

8 years agoMerge "Declare the operator== overload for M128A in the global namespace."
Alex Converse [Wed, 27 Apr 2016 20:33:17 +0000 (20:33 +0000)]
Merge "Declare the operator== overload for M128A in the global namespace."

8 years agoDeclare the operator== overload for M128A in the global namespace.
Alex Converse [Wed, 27 Apr 2016 04:38:04 +0000 (21:38 -0700)]
Declare the operator== overload for M128A in the global namespace.

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

8 years agoMerge changes If45b8931,Ieab273a7
James Zern [Wed, 27 Apr 2016 20:01:33 +0000 (20:01 +0000)]
Merge changes If45b8931,Ieab273a7

* changes:
  vp9_ethread_test: avoid crash when decoder is unavailable
  resize_test: skip VP9 decode checks if it's disabled

8 years agoMerge "TODO cleanup: Remove unused buffer_size"
Vignesh Venkatasubramanian [Wed, 27 Apr 2016 06:17:12 +0000 (06:17 +0000)]
Merge "TODO cleanup: Remove unused buffer_size"

8 years agoTODO cleanup: Remove unused buffer_size
Vignesh Venkatasubramanian [Mon, 25 Apr 2016 20:46:42 +0000 (13:46 -0700)]
TODO cleanup: Remove unused buffer_size

Remove unused variable buffer_size and rename bytes_in_buffer to
buffer_size in webm_read_frame.

Change-Id: I18b6ccbb044cd366610a5e39d83883cc0b174051

8 years agoMerge "webmdec.h: remove stale TODO"
Vignesh Venkatasubramanian [Wed, 27 Apr 2016 05:38:53 +0000 (05:38 +0000)]
Merge "webmdec.h: remove stale TODO"

8 years agoMerge "TODO cleanup: s/EbmlGlobal/WebmOutputContext/"
Vignesh Venkatasubramanian [Wed, 27 Apr 2016 05:37:32 +0000 (05:37 +0000)]
Merge "TODO cleanup: s/EbmlGlobal/WebmOutputContext/"

8 years agovp9_ethread_test: avoid crash when decoder is unavailable
James Zern [Wed, 27 Apr 2016 03:15:15 +0000 (20:15 -0700)]
vp9_ethread_test: avoid crash when decoder is unavailable

Change-Id: If45b8931771d50990b78a5aa860f372eb0430cf4

8 years agoresize_test: skip VP9 decode checks if it's disabled
James Zern [Wed, 27 Apr 2016 02:56:44 +0000 (19:56 -0700)]
resize_test: skip VP9 decode checks if it's disabled

Change-Id: Ieab273a779c44e1c4648642e9f12735b81bd2300

8 years agovp8: Adjust consec_zeromv thresholds for skin detection.
Marco [Tue, 26 Apr 2016 23:02:09 +0000 (16:02 -0700)]
vp8: Adjust consec_zeromv thresholds for skin detection.

Change-Id: Iab5aed7bf86538352232e208bce58c3528cc9a4e

8 years agoMerge "VP9: adjust trellis quant optimization RD parameters"
Hui Su [Tue, 26 Apr 2016 19:12:09 +0000 (19:12 +0000)]
Merge "VP9: adjust trellis quant optimization RD parameters"

8 years agoMerge "VP9: enable trellis quantization optimization for intra blocks"
Hui Su [Tue, 26 Apr 2016 19:11:49 +0000 (19:11 +0000)]
Merge "VP9: enable trellis quantization optimization for intra blocks"

8 years agoMerge "vp9: Fix condition to update consec_zero_mv."
Marco Paniconi [Tue, 26 Apr 2016 18:15:27 +0000 (18:15 +0000)]
Merge "vp9: Fix condition to update consec_zero_mv."

8 years agoMerge "vp9: Simplify the logic in denoiser SSE2 code."
Jacky Chen [Tue, 26 Apr 2016 17:43:48 +0000 (17:43 +0000)]
Merge "vp9: Simplify the logic in denoiser SSE2 code."

8 years agoMerge "Add the 64-bit CPU cycle count utility function"
Yi Luo [Tue, 26 Apr 2016 17:24:21 +0000 (17:24 +0000)]
Merge "Add the 64-bit CPU cycle count utility function"

8 years agoVP9: adjust trellis quant optimization RD parameters
hui su [Mon, 25 Apr 2016 21:23:06 +0000 (14:23 -0700)]
VP9: adjust trellis quant optimization RD parameters

Coding gain:
lowres  0.64%
midres  0.38%
hdres   0.58%

Change-Id: I233fa2a4b24bd1e15091a5f5ef6aff661f3f50ec

8 years agoVP9: enable trellis quantization optimization for intra blocks
hui su [Mon, 25 Apr 2016 19:40:56 +0000 (12:40 -0700)]
VP9: enable trellis quantization optimization for intra blocks

Coding gain:
lowres  0.18%
midres  0.23%
hdres   0.36%

Change-Id: I044c8afbc481fc55b23d440352941071355b0afb

8 years agovp9: Simplify the logic in denoiser SSE2 code.
JackyChen [Tue, 26 Apr 2016 06:52:19 +0000 (23:52 -0700)]
vp9: Simplify the logic in denoiser SSE2 code.

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

8 years agoMerge "Resolve a couple of TODOs in firstpass.c"
Paul Wilkins [Tue, 26 Apr 2016 12:57:08 +0000 (12:57 +0000)]
Merge "Resolve a couple of TODOs in firstpass.c"

8 years agoMerge "Experiment to adapt for net AQ offset."
Paul Wilkins [Tue, 26 Apr 2016 12:56:52 +0000 (12:56 +0000)]
Merge "Experiment to adapt for net AQ offset."

8 years agovp9: Fix consec_zero_mv in skin detection.
Marco [Mon, 25 Apr 2016 22:37:50 +0000 (15:37 -0700)]
vp9: Fix consec_zero_mv in skin detection.

This was missed in https://chromium-review.googlesource.com/#/c/340135/.

Change-Id: Iafc83141f4ff744819bdbde7d239678b215feb9d

8 years agovp9: Fix condition to update consec_zero_mv.
Marco [Mon, 25 Apr 2016 21:34:41 +0000 (14:34 -0700)]
vp9: Fix condition to update consec_zero_mv.

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

8 years agoMerge "vp9: 1 pass vbr: Adjust gf boost/interval on motion level."
Marco Paniconi [Mon, 25 Apr 2016 21:13:56 +0000 (21:13 +0000)]
Merge "vp9: 1 pass vbr: Adjust gf boost/interval on motion level."

8 years agowebmdec.h: remove stale TODO
Vignesh Venkatasubramanian [Mon, 25 Apr 2016 21:05:59 +0000 (14:05 -0700)]
webmdec.h: remove stale TODO

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

8 years agoTODO cleanup: s/EbmlGlobal/WebmOutputContext/
Vignesh Venkatasubramanian [Mon, 25 Apr 2016 20:28:24 +0000 (13:28 -0700)]
TODO cleanup: s/EbmlGlobal/WebmOutputContext/

This brings the struct's name in line with WebmInputContext.

Change-Id: I1a1c7ae60b006ae5de8253177d6d996f5867866d

8 years agoMerge "vp9: Fix a bug where best_sse_mv is not set properly."
Jacky Chen [Mon, 25 Apr 2016 20:00:43 +0000 (20:00 +0000)]
Merge "vp9: Fix a bug where best_sse_mv is not set properly."

8 years agoMerge "vpx_minmax_8x8_neon and test"
Johann Koenig [Mon, 25 Apr 2016 18:58:29 +0000 (18:58 +0000)]
Merge "vpx_minmax_8x8_neon and test"

8 years agovp9: 1 pass vbr: Adjust gf boost/interval on motion level.
Marco [Wed, 20 Apr 2016 22:53:20 +0000 (15:53 -0700)]
vp9: 1 pass vbr: Adjust gf boost/interval on motion level.

Add some adjustment for high motion case.

Change-Id: If700be1ce849c187ed031221ea7cabbc8d91ec2b

8 years agovp9: Fix a bug where best_sse_mv is not set properly.
JackyChen [Mon, 25 Apr 2016 18:44:40 +0000 (11:44 -0700)]
vp9: Fix a bug where best_sse_mv is not set properly.

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

8 years agoMerge "vp9-denoiser: Bugfix and some adjustments for high noise case."
Marco Paniconi [Mon, 25 Apr 2016 17:16:09 +0000 (17:16 +0000)]
Merge "vp9-denoiser: Bugfix and some adjustments for high noise case."

8 years agovp9-denoiser: Bugfix and some adjustments for high noise case.
Marco [Mon, 25 Apr 2016 15:49:08 +0000 (08:49 -0700)]
vp9-denoiser: Bugfix and some adjustments for high noise case.

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

8 years agovp9: Add rc quantity to track amount of low motion in scene.
Marco [Fri, 22 Apr 2016 22:27:53 +0000 (15:27 -0700)]
vp9: Add rc quantity to track amount of low motion in scene.

Use it for now in noise estimation to bypass estimation if
motion level is high.

Change-Id: I033662dc909f2060e4e81abf562a7ad262dc8170

8 years agoMerge "VP9: Do not call vp9_adjust_mask() in vp9_setup_mask()"
Scott LaVarnway [Mon, 25 Apr 2016 13:39:29 +0000 (13:39 +0000)]
Merge "VP9: Do not call vp9_adjust_mask() in vp9_setup_mask()"

8 years agoResolve a couple of TODOs in firstpass.c
paulwilkins [Mon, 25 Apr 2016 11:38:10 +0000 (12:38 +0100)]
Resolve a couple of TODOs in firstpass.c

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

8 years agoExperiment to adapt for net AQ offset.
paulwilkins [Fri, 11 Mar 2016 10:44:16 +0000 (10:44 +0000)]
Experiment to adapt for net AQ offset.

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

8 years agoVP9: Do not call vp9_adjust_mask() in vp9_setup_mask()
Scott LaVarnway [Fri, 22 Apr 2016 20:53:32 +0000 (13:53 -0700)]
VP9: Do not call vp9_adjust_mask() in vp9_setup_mask()

vp9_adjust_mask() is called again in loop_filter_rows().

Change-Id: If52f5339dfa7971c47b12f9e05f87951044d9391

8 years agoMerge "hadamard_test: align src/dst buffers"
James Zern [Fri, 22 Apr 2016 20:03:06 +0000 (20:03 +0000)]
Merge "hadamard_test: align src/dst buffers"

8 years agoMerge "vp9: Some adjustments to denoiser under high-noise condition."
Marco Paniconi [Fri, 22 Apr 2016 18:55:57 +0000 (18:55 +0000)]
Merge "vp9: Some adjustments to denoiser under high-noise condition."

8 years agovp9-noise estimation: Add check on whether skin detection is enabled.
Marco [Fri, 22 Apr 2016 17:35:29 +0000 (10:35 -0700)]
vp9-noise estimation: Add check on whether skin detection is enabled.

Change-Id: Ie80b7a14e3505c40b5742793f84aa6fb9d21dab5

8 years agovp9: Some adjustments to denoiser under high-noise condition.
Marco [Fri, 22 Apr 2016 17:30:21 +0000 (10:30 -0700)]
vp9: Some adjustments to denoiser under high-noise condition.

Change-Id: Ib47d742d07f13c7c39cea85e228c19ae298baf4a

8 years agoMerge "vp9: Move consec_zero_mv from cyclic refresh to cpi struct."
Marco Paniconi [Fri, 22 Apr 2016 17:06:31 +0000 (17:06 +0000)]
Merge "vp9: Move consec_zero_mv from cyclic refresh to cpi struct."

8 years agoAdd the 64-bit CPU cycle count utility function
Yi Luo [Thu, 21 Apr 2016 22:59:05 +0000 (15:59 -0700)]
Add the 64-bit CPU cycle count utility function

Change-Id: Ie87245bbdf5735bc9729199eeb07899d81dbf267

8 years agovp9: Move consec_zero_mv from cyclic refresh to cpi struct.
Marco [Thu, 21 Apr 2016 21:00:26 +0000 (14:00 -0700)]
vp9: Move consec_zero_mv from cyclic refresh to cpi struct.

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

8 years agohadamard_test: align src/dst buffers
James Zern [Fri, 22 Apr 2016 07:06:49 +0000 (00:06 -0700)]
hadamard_test: align src/dst buffers

fixes crash in 32-bit builds

Change-Id: I149e517c940a8cd43643e213bfbe077b6f579090

8 years agovpx_minmax_8x8_neon and test
Johann [Thu, 14 Apr 2016 21:26:32 +0000 (14:26 -0700)]
vpx_minmax_8x8_neon and test

BUG=https://bugs.chromium.org/p/webm/issues/detail?id=1156

Change-Id: Ief0ad8d6255b0ef0f233cda153799e3c72d3dbc6

8 years agoMerge "hadamard 8x8 test"
Johann Koenig [Fri, 22 Apr 2016 04:36:10 +0000 (04:36 +0000)]
Merge "hadamard 8x8 test"

8 years agohadamard 8x8 test
Johann [Fri, 15 Apr 2016 18:35:56 +0000 (11:35 -0700)]
hadamard 8x8 test

The order of the output structure is not currently important.

BUG=https://bugs.chromium.org/p/webm/issues/detail?id=1021

Change-Id: Ibc0006d569675db6c5060c4529f5d9e73f2e96a6

8 years agoMerge "vp9_loopfilter.c : fix / clarify todo"
James Bankoski [Thu, 21 Apr 2016 20:40:40 +0000 (20:40 +0000)]
Merge "vp9_loopfilter.c : fix / clarify todo"

8 years agoMerge "vp9_loopfilter.c: Todo clean up encoder should work like decoder."
James Bankoski [Thu, 21 Apr 2016 20:40:35 +0000 (20:40 +0000)]
Merge "vp9_loopfilter.c: Todo clean up encoder should work like decoder."

8 years agoMerge "vp9_loopfilter.c : todo cleanup"
James Bankoski [Thu, 21 Apr 2016 20:40:24 +0000 (20:40 +0000)]
Merge "vp9_loopfilter.c : todo cleanup"

8 years agovp9_loopfilter.c : fix / clarify todo
Jim Bankoski [Thu, 21 Apr 2016 00:28:52 +0000 (17:28 -0700)]
vp9_loopfilter.c : fix / clarify todo

Change-Id: Ie3ec67a83d1877d3deae9c7922b6899d915aa19e

8 years agovp9_loopfilter.c: Todo clean up encoder should work like decoder.
Jim Bankoski [Thu, 21 Apr 2016 00:24:53 +0000 (17:24 -0700)]
vp9_loopfilter.c: Todo clean up encoder should work like decoder.

Change-Id: I570c6859d6e18cd94ce4a29068477b937489399c

8 years agovp9_encodemb.c: TODO clean up
Jim Bankoski [Thu, 21 Apr 2016 15:48:58 +0000 (08:48 -0700)]
vp9_encodemb.c: TODO clean up

huisu did in nextgen branch -> please try in vp9

Change-Id: I0ff35db07ac38464e0e2858e303be686c03a5d0e

8 years agovp9_loopfilter.c : todo cleanup
Jim Bankoski [Thu, 21 Apr 2016 00:16:25 +0000 (17:16 -0700)]
vp9_loopfilter.c : todo cleanup

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

8 years agoMerge "vp9_loopfilter.c: Todo try inloop calculation."
Yaowu Xu [Thu, 21 Apr 2016 15:38:12 +0000 (15:38 +0000)]
Merge "vp9_loopfilter.c: Todo try inloop calculation."

8 years agovp9_loopfilter.c: Todo try inloop calculation.
Jim Bankoski [Wed, 20 Apr 2016 22:31:46 +0000 (15:31 -0700)]
vp9_loopfilter.c: Todo try inloop calculation.

This is implemented in the decoder already.  Will add a todo for the
encoder.

Change-Id: I5e78c045cb2edb5ba171022aeeb70051a708b916

8 years agovp9: 1 pass vbr: Small fix on reset avg_intersize_gfint on scene cut.
Marco [Wed, 20 Apr 2016 23:30:13 +0000 (16:30 -0700)]
vp9: 1 pass vbr: Small fix on reset avg_intersize_gfint on scene cut.

Change-Id: Ie73d1bbc2c953d010b32097440698f0d060750be

8 years agovp9: 1 pass vbr: small adjustment to active_best factor for inter.
Marco [Wed, 20 Apr 2016 21:22:08 +0000 (14:22 -0700)]
vp9: 1 pass vbr: small adjustment to active_best factor for inter.

Change-Id: I9c68aa9e72969b0ffbf57eb186380ff666745aeb

8 years agoMerge "vp8: Update on sum_diff_threshold."
Marco Paniconi [Wed, 20 Apr 2016 00:20:41 +0000 (00:20 +0000)]
Merge "vp8: Update on sum_diff_threshold."

8 years agovp9: Adjust gf period for high average Q and overshoot.
Marco [Tue, 19 Apr 2016 17:46:20 +0000 (10:46 -0700)]
vp9: Adjust gf period for high average Q and overshoot.

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

8 years agovp8: Update on sum_diff_threshold.
Marco [Tue, 19 Apr 2016 22:51:32 +0000 (15:51 -0700)]
vp8: Update on sum_diff_threshold.

Reset to previous values.

Change-Id: Ia49bafc1d1ec5f657819ffba04d11fd701f315e8

8 years agovp9: 1 pass vbr: More even spacing for gf near key frame.
Marco [Mon, 18 Apr 2016 17:51:07 +0000 (10:51 -0700)]
vp9: 1 pass vbr: More even spacing for gf near key frame.

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

8 years agovp9: Adjustment to active_best_quality for inter_frame, 1 pass vbr.
Marco [Fri, 15 Apr 2016 17:06:18 +0000 (10:06 -0700)]
vp9: Adjustment to active_best_quality for inter_frame, 1 pass vbr.

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

8 years agoMerge "Adjustment to prediction decay."
Paul Wilkins [Mon, 18 Apr 2016 18:47:13 +0000 (18:47 +0000)]
Merge "Adjustment to prediction decay."

8 years agoMerge "VP9: inline vp9_get_intra_inter_context()"
Scott LaVarnway [Fri, 15 Apr 2016 19:06:33 +0000 (19:06 +0000)]
Merge "VP9: inline vp9_get_intra_inter_context()"

8 years agoMerge "Enable vpx_idct32x32_1024_add_neon for neon as well, not only for neon_asm"
Johann Koenig [Fri, 15 Apr 2016 16:00:51 +0000 (16:00 +0000)]
Merge "Enable vpx_idct32x32_1024_add_neon for neon as well, not only for neon_asm"

8 years agoVP9: inline vp9_get_intra_inter_context()
Scott LaVarnway [Fri, 15 Apr 2016 11:58:37 +0000 (04:58 -0700)]
VP9: inline vp9_get_intra_inter_context()

Change-Id: I71366140799b9b39474b9b459082cdb250bd1905

8 years agoEnable vpx_idct32x32_1024_add_neon for neon as well, not only for neon_asm
Martin Storsjo [Thu, 3 Mar 2016 09:55:08 +0000 (11:55 +0200)]
Enable vpx_idct32x32_1024_add_neon for neon as well, not only for neon_asm

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

8 years agoApply 'const' to data not pointer
Johann [Thu, 14 Apr 2016 21:47:16 +0000 (14:47 -0700)]
Apply 'const' to data not pointer

Change-Id: Ic6b695442e319f7582a7ee8e52a47ae3e38c7298

8 years agoMerge "dct32x32_test: s/HAVE_NEON_ASM/HAVE_NEON/"
James Zern [Thu, 14 Apr 2016 02:35:45 +0000 (02:35 +0000)]
Merge "dct32x32_test: s/HAVE_NEON_ASM/HAVE_NEON/"

8 years agoMerge "vp9: Adjustment to scene-cut detection."
Marco Paniconi [Wed, 13 Apr 2016 22:36:28 +0000 (22:36 +0000)]
Merge "vp9: Adjustment to scene-cut detection."

8 years agovp9: Adjustment to scene-cut detection.
Marco [Wed, 13 Apr 2016 18:18:16 +0000 (11:18 -0700)]
vp9: Adjustment to scene-cut detection.

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

8 years agoFix Visual Studio build warning
Yunqing Wang [Wed, 13 Apr 2016 16:09:08 +0000 (09:09 -0700)]
Fix Visual Studio build warning

Fixed warning C4244: '-=' : conversion from 'const double' to 'int',
possible loss of data.

Change-Id: Ic4691346037767b244e7f71248c2f871f92002f3

8 years agodct32x32_test: s/HAVE_NEON_ASM/HAVE_NEON/
James Zern [Wed, 13 Apr 2016 07:05:47 +0000 (00:05 -0700)]
dct32x32_test: s/HAVE_NEON_ASM/HAVE_NEON/

vpx_idct32x32_1024_add_neon is implemented with instrinsics

Change-Id: I072b18248b97ee2634f06b2751ffa2ced85f8e5b

8 years agoMerge changes from topic 'arm64'
James Zern [Wed, 13 Apr 2016 03:11:39 +0000 (03:11 +0000)]
Merge changes from topic 'arm64'

* changes:
  configure: Detect aarch64 toolchains automatically
  configure: Add an arm64-linux-gcc target configuration

8 years agovp9: Adjust threshold for scene-change detection.
Marco [Wed, 13 Apr 2016 01:27:04 +0000 (18:27 -0700)]
vp9: Adjust threshold for scene-change detection.

For 1 pass vbr.

Change-Id: I10b7eefc36d65c30844d205e139515bec7fed6af

8 years agoMerge "Increase active map test coverage from RT speeds 0-5 to 0-8."
Alex Converse [Tue, 12 Apr 2016 22:20:12 +0000 (22:20 +0000)]
Merge "Increase active map test coverage from RT speeds 0-5 to 0-8."

8 years agoMerge "vp9: Fix to active_best for GF/ARF in 1 pass vbr."
Marco Paniconi [Tue, 12 Apr 2016 21:18:41 +0000 (21:18 +0000)]
Merge "vp9: Fix to active_best for GF/ARF in 1 pass vbr."

8 years agoMerge "Step towards making the 2-pass cq mode perceptual"
Debargha Mukherjee [Tue, 12 Apr 2016 19:42:06 +0000 (19:42 +0000)]
Merge "Step towards making the 2-pass cq mode perceptual"

8 years agoIncrease active map test coverage from RT speeds 0-5 to 0-8.
Alex Converse [Tue, 12 Apr 2016 18:14:10 +0000 (11:14 -0700)]
Increase active map test coverage from RT speeds 0-5 to 0-8.

This test takes less than 100 ms for each of speeds 6-8.

Change-Id: Ibbeb3004a2607d25dcbf77cb5314ade87809e059

8 years agovp9: Fix to active_best for GF/ARF in 1 pass vbr.
Marco [Fri, 8 Apr 2016 00:20:16 +0000 (17:20 -0700)]
vp9: Fix to active_best for GF/ARF in 1 pass vbr.

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

8 years agoconfigure: Detect aarch64 toolchains automatically
Martin Storsjo [Thu, 3 Mar 2016 10:00:06 +0000 (12:00 +0200)]
configure: Detect aarch64 toolchains automatically

Change-Id: Icafda81dbc3323fa0afdba5f1c8758e812cc592a

8 years agoconfigure: Add an arm64-linux-gcc target configuration
Martin Storsjo [Thu, 3 Mar 2016 09:53:49 +0000 (11:53 +0200)]
configure: Add an arm64-linux-gcc target configuration

Change-Id: I23efc07572b2406ce5d9283340aef5aee8326280

8 years agopickmode: only cost the skip flag once per prediction block
Alex Converse [Mon, 11 Apr 2016 21:02:34 +0000 (14:02 -0700)]
pickmode: only cost the skip flag once per prediction block

RTC speed 6:
File    Match   Avg     BDRate  Low     Mid     High
OVERALL ✔       -0.040  -0.045  -0.031  -0.084  0.004

Screencast speed 6:
File     Match   Avg     BDRate  Low     Mid     High
OVERALL ✔       1.115   -0.162  0.203   2.470   0.541

Change-Id: I46bbc11c89301015b5d3eac25294c709f23f0897

8 years agoMerge "Remove obsolete segment skip checks from tokenization."
Alex Converse [Mon, 11 Apr 2016 18:58:32 +0000 (18:58 +0000)]
Merge "Remove obsolete segment skip checks from tokenization."

8 years agoAdjustment to prediction decay.
Paul Wilkins [Wed, 30 Mar 2016 12:29:25 +0000 (13:29 +0100)]
Adjustment to prediction decay.

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

8 years agoMerge "Adjust motion component of prediction decay."
Paul Wilkins [Mon, 11 Apr 2016 10:27:29 +0000 (10:27 +0000)]
Merge "Adjust motion component of prediction decay."

8 years agoMerge "Trap very short arf group just before a kf."
Paul Wilkins [Mon, 11 Apr 2016 10:27:12 +0000 (10:27 +0000)]
Merge "Trap very short arf group just before a kf."