]> granicus.if.org Git - libvpx/log
libvpx
7 years agovp9: speed >= 8: Skip uv variance in model_rd_sb_y_large
Jerome Jiang [Fri, 21 Apr 2017 17:10:05 +0000 (10:10 -0700)]
vp9: speed >= 8: Skip uv variance in model_rd_sb_y_large

For speed >= 8 and color_sensitivity not set, skip the transform
skipping test in UV planes.
Add a new condition to check noise level to skip chroma check
for speed >= 8 if y_sad is high.

1~2% speedup on ARM for speed 8.

Borg tests show neutral results in both rtc and rtc_derf.

Change-Id: Idecd3ff6e28c97757a43bb6f3a7082c85f72109c

7 years agoMerge "vp9: Non-rd pickmode: Avoid computation duplication."
Jerome Jiang [Fri, 21 Apr 2017 00:51:46 +0000 (00:51 +0000)]
Merge "vp9: Non-rd pickmode:  Avoid computation duplication."

7 years agovp9: Non-rd pickmode: Avoid computation duplication.
Jerome Jiang [Thu, 20 Apr 2017 17:57:02 +0000 (10:57 -0700)]
vp9: Non-rd pickmode:  Avoid computation duplication.

In non-rd pickmode (speed >= 5), avoid duplication of computations in
model_rd_for_sb_y when the speed feature use_simple_block_yrd is
enabled (or for high bitdepth build under certain conditions).

QVGA, VGA and HD have 1.23%, 2.68% and 1.7% speedup on ARM for speed 8,
respectively.

Encoding results are bitexact for speed >= 5.

Change-Id: I3f9130810c21439f5ad7e159e21cb2243dcd05f1

7 years agovp9: 1 pass SVC: Fix comment and condition for up-sampling reference.
Marco [Thu, 20 Apr 2017 21:13:57 +0000 (14:13 -0700)]
vp9: 1 pass SVC: Fix comment and condition for up-sampling reference.

No change in behavior.

Change-Id: I218fb30289091da623acb23324027435b8510d0e

7 years agoMerge "Only allow allow_exhaustive_searches for FC_GRAPHICS_ANIMATION content"
Yunqing Wang [Thu, 20 Apr 2017 19:57:46 +0000 (19:57 +0000)]
Merge "Only allow allow_exhaustive_searches for FC_GRAPHICS_ANIMATION content"

7 years agoMerge "vp9: Re-enable SVC datarate tests."
Marco Paniconi [Thu, 20 Apr 2017 19:53:19 +0000 (19:53 +0000)]
Merge "vp9: Re-enable SVC datarate tests."

7 years agovp9: Re-enable SVC datarate tests.
Marco [Wed, 19 Apr 2017 18:12:42 +0000 (11:12 -0700)]
vp9: Re-enable SVC datarate tests.

Re-enable the SVC tests, wrap the non-zero expectation
in GetMismatchFrames around #if CONFIG_VP9_DECODER.

Change-Id: I0e8a2d78b868c32f18fe597540f397d3a1b303b5

7 years agovp9: SVC: Redefine the source downsample filter choice.
Marco [Thu, 20 Apr 2017 16:50:16 +0000 (09:50 -0700)]
vp9: SVC: Redefine the source downsample filter choice.

Rename the source downsampling filter, and define it
per spatial layers. Used 1 pass CBR SVC.

Change-Id: I8135f2ab89c535c53429b9c58b586f746bb668c7

7 years agoppc: Add the intra predictor tests
Luca Barbato [Tue, 18 Apr 2017 23:37:57 +0000 (23:37 +0000)]
ppc: Add the intra predictor tests

Change-Id: Idea15b916044ab3d8e74519337880a484ecfd87e

7 years agoppc: h predictor 8x8
Luca Barbato [Tue, 18 Apr 2017 22:55:53 +0000 (22:55 +0000)]
ppc: h predictor 8x8

Slightly faster with the current compiler.

Change-Id: Iae225fac08395eb430c97a2abec69c60f5cf5c47

7 years agoppc: d63 predictor 8x8
Luca Barbato [Tue, 11 Apr 2017 23:18:35 +0000 (01:18 +0200)]
ppc: d63 predictor 8x8

10x faster.

Change-Id: I7cedbf4df2ce7df5b6f1108b11815d088fdb9ba8

7 years agoppc: tm predictor 4x4
Luca Barbato [Sun, 9 Apr 2017 15:07:03 +0000 (15:07 +0000)]
ppc: tm predictor 4x4

Slightly faster.

Change-Id: I0ca43f309b3d9b50435d69bd5be64b53a99bd191

7 years agoppc: h predictor 4x4
Luca Barbato [Sun, 9 Apr 2017 13:44:41 +0000 (13:44 +0000)]
ppc: h predictor 4x4

2x faster.

Change-Id: I0583dec353299c6797401b646099f18db4e0420d

7 years agoppc: dc predictor 8x8
Luca Barbato [Sun, 9 Apr 2017 13:05:09 +0000 (13:05 +0000)]
ppc: dc predictor 8x8

Slightly faster, the other dc predictors cannot be faster since
the computation speedup is overwhelmed by the time spent reading
dst to write just the 8x8 part.

Change-Id: I94a0b50500adf8b7b6bb919dbf5c7adf5b9fba66

7 years agoppc: d45 predictor 8x8
Luca Barbato [Sun, 9 Apr 2017 11:07:22 +0000 (11:07 +0000)]
ppc: d45 predictor 8x8

11x faster.

Change-Id: I5b8f39213ee1f5260724fc254e3fb5c462435798

7 years agoppc: d63 predictor 32x32
Luca Barbato [Sun, 9 Apr 2017 00:09:56 +0000 (00:09 +0000)]
ppc: d63 predictor 32x32

About 10x faster.

Change-Id: If7d0645f75c5d7deb9751edd0bf47e2f9068e9e7

7 years agoppc: d63 predictor 16x16
Luca Barbato [Sun, 9 Apr 2017 00:09:56 +0000 (00:09 +0000)]
ppc: d63 predictor 16x16

About 18x faster.

Change-Id: Id043bf76c011e03e992085bb5e20f330d3e98cd4

7 years agoppc: d45 predictor 32x32
Luca Barbato [Sat, 8 Apr 2017 22:41:41 +0000 (22:41 +0000)]
ppc: d45 predictor 32x32

About 12x faster.

Change-Id: I22c150256aefb4941861ab1f6c17d554fb694bed

7 years agoppc: d45 predictor 16x16
Luca Barbato [Sat, 8 Apr 2017 22:41:41 +0000 (22:41 +0000)]
ppc: d45 predictor 16x16

About 16x faster.

Change-Id: Ie5469fb32d5fd11bb6cb06318cea475d8a5b00b9

7 years agoppc: dc predictor 32x32
Luca Barbato [Sat, 8 Apr 2017 02:55:33 +0000 (02:55 +0000)]
ppc: dc predictor 32x32

10x and 5x faster.

Change-Id: I7913c58c768334d818f541a5e219f1035791eeaf

7 years agoppc: dc top and left predictor 32x32
Luca Barbato [Sat, 8 Apr 2017 02:55:33 +0000 (02:55 +0000)]
ppc: dc top and left predictor 32x32

6x faster.

Change-Id: I717995b4056e5579c68191d11b495372971fe1ae

7 years agoppc: dc top and left predictor 16x16
Luca Barbato [Sat, 8 Apr 2017 02:55:33 +0000 (02:55 +0000)]
ppc: dc top and left predictor 16x16

13x faster.

Change-Id: I1771ac39fda599153f933cb3f0506c9f97a6cbe6

7 years agoppc: dc_128 predictor 32x32
Luca Barbato [Sat, 8 Apr 2017 00:39:24 +0000 (00:39 +0000)]
ppc: dc_128 predictor 32x32

6x faster.

Change-Id: I1da8f51b4262871cb98f0aa03ccda41b0ac2b08b

7 years agoppc: dc_128 predictor 16x16
Luca Barbato [Sat, 8 Apr 2017 00:26:54 +0000 (00:26 +0000)]
ppc: dc_128 predictor 16x16

20x faster.

Change-Id: I05f0deb2d38ae7966eae6b71fbc0aa51880e5709

7 years agoppc: tm predictor 32x32
Luca Barbato [Fri, 7 Apr 2017 14:49:00 +0000 (14:49 +0000)]
ppc: tm predictor 32x32

About 8x faster.

Change-Id: I9bad827ccbdf47ec95406e961c74ac2ff45f80cf

7 years agoMerge changes I1f5a3752,I95123051,I3bb724e0,Ie81077fa,Ic80f3c05, ...
James Zern [Thu, 20 Apr 2017 02:45:44 +0000 (02:45 +0000)]
Merge changes I1f5a3752,I95123051,I3bb724e0,Ie81077fa,Ic80f3c05, ...

* changes:
  ppc: tm predictor 16x16
  ppc: tm predictor 8x8
  ppc: horizontal predictor 32x32
  ppc: horizontal predictor 16x16
  ppc: vertical intrapred 16x16 and 32x32
  configure: Workaround clang not enabling altivec on -mvsx
  configure: Match power*64* as ppc64

7 years agoOnly allow allow_exhaustive_searches for FC_GRAPHICS_ANIMATION content
Yunqing Wang [Wed, 19 Apr 2017 23:32:59 +0000 (16:32 -0700)]
Only allow allow_exhaustive_searches for FC_GRAPHICS_ANIMATION content

The allow_exhaustive_searches feature improves the encoding quality
of FC_GRAPHICS_ANIMATION content a lot. For non-FC_GRAPHICS_ANIMATION
content, the quality test result is almost neutral. This patch makes
this feature to be used only for FC_GRAPHICS_ANIMATION content.

The motivation of doing that is to make this feature no longer adaptive,
which will be implemented in the following patch.

Change-Id: Ic911df6dd757402b6480789cc247801e99840369

7 years agoMerge changes I9e18a73b,Ie47c8cd4
Linfeng Zhang [Wed, 19 Apr 2017 23:55:57 +0000 (23:55 +0000)]
Merge changes I9e18a73b,Ie47c8cd4

* changes:
  Clean CONVERT_TO_BYTEPTR/SHORTPTR in convolve
  Create CAST_TO_BYTEPTR/SHORTPTR

7 years agoClean CONVERT_TO_BYTEPTR/SHORTPTR in convolve
Linfeng Zhang [Thu, 6 Apr 2017 00:54:42 +0000 (17:54 -0700)]
Clean CONVERT_TO_BYTEPTR/SHORTPTR in convolve

Replace by CAST_TO_BYTEPTR/SHORTPTR.
The rule is: if a short ptr is casted to a byte ptr, any offset
operation on the byte ptr must be doubled. We do this by casting to
short ptr first, adding offset, then casting back to byte ptr.

BUG=webm:1388

Change-Id: I9e18a73ba45ddae58fc9dae470c0ff34951fe248

7 years agoMerge "vp9: Add phase to get averaging filter for 1:2 downsampling."
Marco Paniconi [Wed, 19 Apr 2017 15:27:55 +0000 (15:27 +0000)]
Merge "vp9: Add phase to get averaging filter for 1:2 downsampling."

7 years agovp9: Fix the disabling of a SVC 3TL datarate test.
Marco [Wed, 19 Apr 2017 14:59:59 +0000 (07:59 -0700)]
vp9: Fix the disabling of a SVC 3TL datarate test.

Change-Id: Ib42d23ab5ee39ab3c85e1d9a84e36249e59fe74e

7 years agovp9: Add phase to get averaging filter for 1:2 downsampling.
Marco [Fri, 14 Apr 2017 00:19:06 +0000 (17:19 -0700)]
vp9: Add phase to get averaging filter for 1:2 downsampling.

The scaling filter with zero shift will give sub-sampling for
2x downsampling. Allow for a phase shift to get an averaging filter.

Usage is for source scaling in 1 pass SVC mode for 1:2 downscale.
Reduces aliasing in downsampled image.

Keep the phase to 0/off for now.

Change-Id: Ic547ea0748d151b675f877527e656407fcf4d51e

7 years agoppc: tm predictor 16x16
Luca Barbato [Fri, 7 Apr 2017 14:49:00 +0000 (14:49 +0000)]
ppc: tm predictor 16x16

About 10x faster.

Change-Id: I1f5a3752d346459df3b45f92963208bf3e520f06

7 years agoppc: tm predictor 8x8
Luca Barbato [Fri, 7 Apr 2017 14:49:00 +0000 (14:49 +0000)]
ppc: tm predictor 8x8

About 5x faster.

Change-Id: I951230517f49c0dca9ac9eac2efa8916a303b85a

7 years agoppc: horizontal predictor 32x32
Luca Barbato [Fri, 7 Apr 2017 14:49:00 +0000 (14:49 +0000)]
ppc: horizontal predictor 32x32

About 5x faster.

Change-Id: I3bb724e07baffd901aa2d0f65060ba48882cc9b8

7 years agoppc: horizontal predictor 16x16
Luca Barbato [Fri, 7 Apr 2017 14:49:00 +0000 (14:49 +0000)]
ppc: horizontal predictor 16x16

About 10x faster.

Change-Id: Ie81077fa32ad214cdb46bdcb0be4e9e2c7df47c2

7 years agoppc: vertical intrapred 16x16 and 32x32
Luca Barbato [Fri, 7 Apr 2017 13:50:12 +0000 (13:50 +0000)]
ppc: vertical intrapred 16x16 and 32x32

Change-Id: Ic80f3c050cfbe7697e81a311b4edaaa597b85cab

7 years agoconfigure: Workaround clang not enabling altivec on -mvsx
Luca Barbato [Tue, 18 Apr 2017 18:31:18 +0000 (18:31 +0000)]
configure: Workaround clang not enabling altivec on -mvsx

The flag `-mvsx` implies `-maltivec`.

Change-Id: I7544553eba131a533467b387f8bf329d57f5af5c

7 years agoconfigure: Match power*64* as ppc64
Luca Barbato [Fri, 7 Apr 2017 13:14:35 +0000 (13:14 +0000)]
configure: Match power*64* as ppc64

Change-Id: Ie640dff50a5db935bb57c5a2570b423ce8946f2c

7 years agoCreate CAST_TO_BYTEPTR/SHORTPTR
Linfeng Zhang [Thu, 6 Apr 2017 00:40:12 +0000 (17:40 -0700)]
Create CAST_TO_BYTEPTR/SHORTPTR

They will replace CONVERT_TO_BYTEPTR/SHORTPTR module by module.

BUG=webm:1388

Change-Id: Ie47c8cd4897696481b9cbbf9e2d439dc22dc85ec

7 years agovp9: Disable some SVC tests for now.
Marco [Tue, 18 Apr 2017 16:43:32 +0000 (09:43 -0700)]
vp9: Disable some SVC tests for now.

Disable the 1 pass CBR SVC tests with temporal_layers > 1.
Issue with the commit 863f860, which will cause encoder/decoder
mismatch due to skipping encoder loopfilter for non-reference frames.

Will re-enable the tests when fixed.

Change-Id: I74918a0045a17976b069c4be63fbeb921974df0d

7 years agovp9: Add key_frame condition to is_reference check for loopfilter.
Marco [Mon, 17 Apr 2017 21:36:11 +0000 (14:36 -0700)]
vp9: Add key_frame condition to is_reference check for loopfilter.

This condiiton is not needed as key_frame should set the refresh
of the reference frames, but good to have for clarity in condition.

Change-Id: Icf9838e7e4f0ff5cf0a9562ae3b5d6c7e6f78702

7 years agoMerge "re-enable vpx_comp_avg_pred_sse2"
Johann Koenig [Mon, 17 Apr 2017 22:07:34 +0000 (22:07 +0000)]
Merge "re-enable vpx_comp_avg_pred_sse2"

7 years agoRevert "Revert "vp9: Avoid encoder loopfilter for non-reference frames.""
Marco Paniconi [Mon, 17 Apr 2017 18:00:09 +0000 (18:00 +0000)]
Revert "Revert "vp9: Avoid encoder loopfilter for non-reference frames.""

This reverts commit e9b7f98c56b3b9c99a60eb41b83bf8346b3ad25f.

Reason for revert:
Commit d578bdad fixes the issue (encoder/decoder mismatch
in 3TL datarate test) that causes the original revert.

Original change's description:
> Revert "vp9: Avoid encoder loopfilter for non-reference frames."
>
> This reverts commit 863f860bfcf3bdc26eeecb299aa481d0f63d11ac.
>
> This causes encoder / decoder mismatches in various
> VP9/DatarateTestVP9Large.BasicRateTargeting3TemporalLayers tests
>
> BUG=webm:1408
>
> Change-Id: Ic200c39d7ed9c0b0247ef562f5d6f7b2625f7e14
>

TBR=jzern@google.com,marpan@google.com,builds@webmproject.org,jianj@google.com
BUG=webm:1408

Change-Id: Ifeb81460856d1d56482d4e0477a70ee98f8bfaa6

7 years agovp9: Datarate test: modify frame flags for 3 TL.
Marco [Mon, 17 Apr 2017 16:19:03 +0000 (09:19 -0700)]
vp9: Datarate test: modify frame flags for 3 TL.

Modify the frame flags to update the ARF on top layer,
for the tests:
VP9/DatarateTestVP9Large.BasicRateTargeting3TemporalLayers
VP9/DatarateTestVP9Large.BasicRateTargeting3TemporalLayersFrameDropping

This is needed to fix the encode/decoder mismatches caused by 863f860,
and removed in the revert e9b7f98.

Change-Id: I6b9fecfdd17315fc0179e29949338c77636026c0

7 years agore-enable vpx_comp_avg_pred_sse2
Johann [Mon, 17 Apr 2017 15:38:02 +0000 (08:38 -0700)]
re-enable vpx_comp_avg_pred_sse2

Buffers on 32 bit x86 builds only guaranteed 8 byte alignment. Fixed
with "AvgPred test: use aligned buffers" and "sad avg: align
intermediate buffer"

Also re-enable asserts on the C version.

BUG=webm:1390

Change-Id: I93081f1b0002a352bb0a3371ac35452417fa8514

7 years agoMerge "AvgPred test: use aligned buffers"
Johann Koenig [Mon, 17 Apr 2017 15:36:41 +0000 (15:36 +0000)]
Merge "AvgPred test: use aligned buffers"

7 years agosad avg: align intermediate buffer
Johann [Fri, 14 Apr 2017 21:37:58 +0000 (14:37 -0700)]
sad avg: align intermediate buffer

comp_avg_pred has started declaring a requirement for aligned buffers.

BUG=webm:1390

Change-Id: Idaf6667498ea343e8d49b32bc9d8b9d0aa43ef5c

7 years agoMerge "Add AVX2 optimization to copy/avg functions"
James Zern [Sat, 15 Apr 2017 00:26:08 +0000 (00:26 +0000)]
Merge "Add AVX2 optimization to copy/avg functions"

7 years agoAdd AVX2 optimization to copy/avg functions
Yi Luo [Tue, 28 Mar 2017 22:30:07 +0000 (15:30 -0700)]
Add AVX2 optimization to copy/avg functions

Change-Id: Ibcef70e4fead74e2c2909330a7044a29381a8074

7 years agoMerge "Disable vpx_comp_avg_pred_sse2"
Johann Koenig [Fri, 14 Apr 2017 22:01:38 +0000 (22:01 +0000)]
Merge "Disable vpx_comp_avg_pred_sse2"

7 years agoAvgPred test: use aligned buffers
Johann [Fri, 14 Apr 2017 19:44:06 +0000 (12:44 -0700)]
AvgPred test: use aligned buffers

BUG=webm:1390

Change-Id: Idb6d1ce119a09c5e7c9f3c58bbbae3de63463d1d

7 years agoRevert "vp9: Avoid encoder loopfilter for non-reference frames."
James Zern [Fri, 14 Apr 2017 18:28:25 +0000 (11:28 -0700)]
Revert "vp9: Avoid encoder loopfilter for non-reference frames."

This reverts commit 863f860bfcf3bdc26eeecb299aa481d0f63d11ac.

This causes encoder / decoder mismatches in various
VP9/DatarateTestVP9Large.BasicRateTargeting3TemporalLayers tests

BUG=webm:1408

Change-Id: Ic200c39d7ed9c0b0247ef562f5d6f7b2625f7e14

7 years agoMerge "vp9: SVC: fix to allow use_base_mv to be used for 3 layers."
Marco Paniconi [Fri, 14 Apr 2017 17:12:58 +0000 (17:12 +0000)]
Merge "vp9: SVC: fix to allow use_base_mv to be used for 3 layers."

7 years agoDisable vpx_comp_avg_pred_sse2
Johann [Fri, 14 Apr 2017 15:24:59 +0000 (08:24 -0700)]
Disable vpx_comp_avg_pred_sse2

Failures on windows:
unknown file: error: SEH exception with code 0xc0000005 thrown in the
test body.

Alignment check errors on linux:
test_libvpx: ../libvpx/vpx_dsp/variance.c:230: void
vpx_comp_avg_pred_c(uint8_t *, const uint8_t *, int, int, const uint8_t
*, int): Assertion `((intptr_t)comp_pred & 0xf) == 0' failed.

BUG=webm:1390

Change-Id: I5eed5381c0f1a8fe594a128eb415e77232f544ea

7 years agoMerge "vpx_comp_avg_pred: sse2 optimization"
Johann Koenig [Fri, 14 Apr 2017 04:10:55 +0000 (04:10 +0000)]
Merge "vpx_comp_avg_pred: sse2 optimization"

7 years agovp9: SVC: fix to allow use_base_mv to be used for 3 layers.
Marco [Fri, 14 Apr 2017 00:45:55 +0000 (17:45 -0700)]
vp9: SVC: fix to allow use_base_mv to be used for 3 layers.

Allow use_base_mv to be used for 3 spatial layers where
base is 4x4 scale from the top layer.

Change-Id: If6641baf8b8e4d0fd5dc67619d873c6d75065f43

7 years agoMerge "vp9: Avoid encoder loopfilter for non-reference frames."
Marco Paniconi [Fri, 14 Apr 2017 00:45:41 +0000 (00:45 +0000)]
Merge "vp9: Avoid encoder loopfilter for non-reference frames."

7 years agovp9: 1 pass VBR: Fix to rate control at low min-q.
Marco [Thu, 13 Apr 2017 00:06:03 +0000 (17:06 -0700)]
vp9: 1 pass VBR: Fix to rate control at low min-q.

Fix to avoid getting stuck at very low Q even
though content is changing, which can happen for --min-q=0.

Fix is to more aggressively increase active_worst_quality
when detecting significant rate_deviation at very low Q.

Change will only affect 1 pass VBR for --min-q < 4, so no
change in ytlive metrics for --min-q >= 4.

Change-Id: I4dd77dd7c08a30a4390da0ff2c8bda6fccfa76d7

7 years agovp9: Avoid encoder loopfilter for non-reference frames.
Marco [Tue, 11 Apr 2017 23:17:18 +0000 (16:17 -0700)]
vp9: Avoid encoder loopfilter for non-reference frames.

Useful for SVC, where the top layer enhancement frames may
not update any reference buffers, as is the case for the
patterns in the 1 pass CBR SVC when #temporal_layers > 1.

~3% encoder speedup for SVC patterns with temporal layers
in 1 pass CBR mode.

Updated the SVC datarate tests for the mismatch frames.
Set the frame-dropper off in some tests with #temporal_layers > 1
so we can correctly set #mismatch frames. Adjusted rate target
threshold for tests where frame-dropper was turned off.

Change-Id: Ia0c142f02100be0fed61cd2049691be9c59d6793

7 years agovpx_comp_avg_pred: sse2 optimization
Johann [Thu, 23 Mar 2017 21:54:48 +0000 (14:54 -0700)]
vpx_comp_avg_pred: sse2 optimization

Provides over 15x speedup for width > 8.

Due to smaller loads and shifting for width == 8 it gets about 8x
speedup.

For width == 4 it's only about 4x speedup because there is a lot of
shuffling and shifting to get the data properly situated.

BUG=webm:1390

Change-Id: Ice0b3dbbf007be3d9509786a61e7f35e94bdffa8

7 years agoFix an integer overflow in vp9_mcomp.c
Yunqing Wang [Mon, 10 Apr 2017 17:57:41 +0000 (10:57 -0700)]
Fix an integer overflow in vp9_mcomp.c

The MV unit test revealed an integer overflow issue in vp9_mcomp.c.
This was caused if the MV was very large. In mv_err_cost(), when
mv->row = 8184, mv->col = 8184 and ref_mv is 0, mv_cost = 34363
and error_per_bit = 132412, causing the overflow.

BUG=webm:1406

Change-Id: I35f8299f22f9bee39cd9153d7b00d0993838845e

7 years agoMerge "vp9: speed >= 8: Adjust speed settings on ARM."
Jerome Jiang [Tue, 11 Apr 2017 00:45:20 +0000 (00:45 +0000)]
Merge "vp9: speed >= 8: Adjust speed settings on ARM."

7 years agovp9: speed >= 8: Adjust speed settings on ARM.
Jerome Jiang [Mon, 10 Apr 2017 20:55:35 +0000 (13:55 -0700)]
vp9: speed >= 8: Adjust speed settings on ARM.

Set adaptive_rd_thresh to 2 when simple block yrd is not used.

Fix regression caused by computing y sad without
int_pro_motion_estimation on low res motion clips.

Overall 0.07% quality loss on rtc_derf.

Change only affects low res on speed 8.

Change-Id: Ic6a188a56529f1034d6431005fb4b0e24e8a7e27

7 years agovp9: 1 pass CBR: avoid nonrd_pick_partition on segment.
Marco [Mon, 10 Apr 2017 21:56:46 +0000 (14:56 -0700)]
vp9: 1 pass CBR: avoid nonrd_pick_partition on segment.

For speed 5, 1 pass CBR: Don't use the nonrd_pick_partition
on the segment, rather use choose_partitioning followed by
nonrd_select_partition (as is done on base segment).

Little/no quality loss on RTC and RTC_derf (< 0.3%),
speedup of at least 5%.

Change-Id: I5273d5f950e60adf5e437b4ca8c4f63964641e83

7 years agoMerge "vp9: Fix to noise estimation for temporal denoising."
Marco Paniconi [Fri, 7 Apr 2017 17:13:21 +0000 (17:13 +0000)]
Merge "vp9: Fix to noise estimation for temporal denoising."

7 years agoMerge "VP9 motion vector unit test"
Yunqing Wang [Fri, 7 Apr 2017 16:46:22 +0000 (16:46 +0000)]
Merge "VP9 motion vector unit test"

7 years agovp9: Fix to noise estimation for temporal denoising.
Marco [Thu, 6 Apr 2017 23:35:01 +0000 (16:35 -0700)]
vp9: Fix to noise estimation for temporal denoising.

If the noise estimation is avoided due to large motion,
the last_source for denoising should still be updated.

Change-Id: I67155ea7dbe9ac2785978e64a27bdafd7d57aac0

7 years agovp9: Adjust consec_zeromv threshold for aq-mode=3.
Marco [Fri, 7 Apr 2017 15:52:54 +0000 (08:52 -0700)]
vp9: Adjust consec_zeromv threshold for aq-mode=3.

To reduce refresh on partial super-blocks on boundary,
for noisy input. Reduces some artifacts on noisy input.

Change-Id: I10b5808a296874e08c7f378b3df58466591d8dbe
Edit

7 years agoMerge changes from topic 'Wshorten'
James Zern [Fri, 7 Apr 2017 07:32:14 +0000 (07:32 +0000)]
Merge changes from topic 'Wshorten'

* changes:
  configure: enable -Wshorten-64-to-32 for hbd
  vp9_encodeframe: resolve -Wshorten-64-to-32 in hbd
  Resolve -Wshorten-64-to-32 in highbd variance.

7 years agoMerge "Fix compile warnings with enable-internal-stats flag."
Jerome Jiang [Fri, 7 Apr 2017 03:34:54 +0000 (03:34 +0000)]
Merge "Fix compile warnings with enable-internal-stats flag."

7 years agoFix compile warnings with enable-internal-stats flag.
Jerome Jiang [Thu, 6 Apr 2017 20:07:53 +0000 (13:07 -0700)]
Fix compile warnings with enable-internal-stats flag.

BUG=webm:1402

Change-Id: Ibe9ecb1b559a4b989f6ccedbd097e369f6edde1e

7 years agovp9; Move the denoising condition for speed 5.
Marco [Thu, 6 Apr 2017 17:25:55 +0000 (10:25 -0700)]
vp9; Move the denoising condition for speed 5.

Move the condition for effectively disabling the denoising
for speed 5 into the vp9_denoiser_denoise().

This is cleaner, and also moving the condition into vp9_denoiser_denoise
will keep the denoiser buffer updated with the current source.
This allows for more consistent behavior if speed is changed midstream.

Change-Id: Ia001f591c56e454bf724c3ae73c024badb183ef8

7 years agoMerge "vp9: speed 8: Compute y sad without int_pro_motion_estimation."
Jerome Jiang [Thu, 6 Apr 2017 02:57:16 +0000 (02:57 +0000)]
Merge "vp9: speed 8: Compute y sad without int_pro_motion_estimation."

7 years agoMerge "Refactor: Clean memory allocation for copy partition."
Jerome Jiang [Thu, 6 Apr 2017 02:57:07 +0000 (02:57 +0000)]
Merge "Refactor: Clean memory allocation for copy partition."

7 years agoVP9 motion vector unit test
Yunqing Wang [Fri, 31 Mar 2017 17:39:57 +0000 (10:39 -0700)]
VP9 motion vector unit test

To prevent the motion vector out of range bug, added a motion vector unit
test in VP9. In the 4k video encoding, always forced to use extreme motion
vectors and also encouraged to use INTER modes. In the decoding, checked if
the motion vector was valid, and also checked the encoder/decoder mismatch.

The tests showed that this unit test could reveal the issue we saw before.

Change-Id: I0a880bd847dad8a13f7fd2012faf6868b02fa3b4

7 years agoconfigure: enable -Wshorten-64-to-32 for hbd
James Zern [Wed, 5 Apr 2017 04:08:51 +0000 (21:08 -0700)]
configure: enable -Wshorten-64-to-32 for hbd

Change-Id: I4041f3fc4aabc7a2251c44c75a477b659284c3cf

7 years agovp9_encodeframe: resolve -Wshorten-64-to-32 in hbd
James Zern [Wed, 5 Apr 2017 04:04:18 +0000 (21:04 -0700)]
vp9_encodeframe: resolve -Wshorten-64-to-32 in hbd

vp9_high_get_sby_perpixel_variance the variance operated on in is
already in 32-bits

Change-Id: I97006eb9c08dbd0f88ee35e1a1ca205737508296

7 years agoResolve -Wshorten-64-to-32 in highbd variance.
James Zern [Wed, 5 Apr 2017 03:37:17 +0000 (20:37 -0700)]
Resolve -Wshorten-64-to-32 in highbd variance.

For 8-bit the subtrahend is small enough to fit into uint32_t.

This is the same that was done for:
c0241664a Resolve -Wshorten-64-to-32 in variance.

For 10/12-bit apply:
63a37d16f Prevent negative variance

Change-Id: Iab35e3f3f269035e17c711bd6cc01272c3137e1d

7 years agovp9: speed 8: Compute y sad without int_pro_motion_estimation.
Jerome Jiang [Wed, 5 Apr 2017 00:47:03 +0000 (17:47 -0700)]
vp9: speed 8: Compute y sad without int_pro_motion_estimation.

Little change in overall PSNR in rtc. 2-4% speedup on VGA on ARM.

Change-Id: I3395806d7afd456deacd4077c330adca13ab0645

7 years agoMerge "vp9: Temporal denoising: avoid denoising for speed <= 5."
Marco Paniconi [Thu, 6 Apr 2017 00:25:44 +0000 (00:25 +0000)]
Merge "vp9: Temporal denoising: avoid denoising for speed <= 5."

7 years agovp9: Temporal denoising: avoid denoising for speed <= 5.
Marco [Wed, 5 Apr 2017 23:38:33 +0000 (16:38 -0700)]
vp9: Temporal denoising: avoid denoising for speed <= 5.

Temporal denoiser runs in non-rd pickmode, so it is only used
for speed >= 5. Regression exists for speed 5, due to use of
reference_partition (which use non-rd pickmode for partitioning).
Avoid denoising for now at speed 5.

Change-Id: I74a74d2e1404d7cfd33dcf4ec06dd2e503256cf0

7 years agoRefactor: Clean memory allocation for copy partition.
Jerome Jiang [Fri, 31 Mar 2017 21:24:03 +0000 (14:24 -0700)]
Refactor: Clean memory allocation for copy partition.

Move the memory allocation from setting speed features.

Change-Id: I2e89dfaeb46daee63effe5a5df62feed732aa990

7 years agoUpdate 32x32 high bitdepth idct NEON optimization
Linfeng Zhang [Wed, 5 Apr 2017 21:41:35 +0000 (14:41 -0700)]
Update 32x32 high bitdepth idct NEON optimization

Preparation of CONVERT_TO_BYTEPTR/SHORTPTR clean up.

BUG=webm:1388

Change-Id: I928d30a5698023bb90888d783cf81c51ec183760

7 years agovp9: Remove legacy comments for avg_source_sad.
Jerome Jiang [Tue, 4 Apr 2017 23:28:27 +0000 (16:28 -0700)]
vp9: Remove legacy comments for avg_source_sad.

Change-Id: Ia6e8614535a097f17f37fc382cef8e22e03b70f6

7 years agovp9: Adjust condition of golden update with cyclic refresh.
Marco [Tue, 4 Apr 2017 16:53:13 +0000 (09:53 -0700)]
vp9: Adjust condition of golden update with cyclic refresh.

Base the low_content_frame metric on the motion vectors,
and adjust the logic for preventing golden update.

Small change in behavior: small positive gain (~0.2-1%) on clips
with high activity.

Change-Id: I0b861c8e9666cd82b45cde5ee57ee8a1e5ab453c

7 years agovp9: 1 pass CBR: cleanup to cyclic refresh.
Marco [Wed, 29 Mar 2017 23:57:53 +0000 (16:57 -0700)]
vp9: 1 pass CBR: cleanup to cyclic refresh.

Code cleanup: merged two functions that were doing postencode
update for cylic refresh, remove some unused code and fix comments.

No change in behavior.

Change-Id: I9be0d7e346d34dec29bf4e5bb380a7bf81c8480a

7 years agoMerge "Fix for out of range motion vector bug in sub-pel motion estimation"
Yunqing Wang [Mon, 3 Apr 2017 18:27:56 +0000 (18:27 +0000)]
Merge "Fix for out of range motion vector bug in sub-pel motion estimation"

7 years agoMerge "vp9: SVC: Fix issue with artifact for svc-denoising."
Marco Paniconi [Mon, 3 Apr 2017 16:23:25 +0000 (16:23 +0000)]
Merge "vp9: SVC: Fix issue with artifact for svc-denoising."

7 years agoFix for out of range motion vector bug in sub-pel motion estimation
Ranjit Kumar Tulabandu [Wed, 22 Mar 2017 14:08:59 +0000 (19:38 +0530)]
Fix for out of range motion vector bug in sub-pel motion estimation

BUG=webm:1397

(yunqingwang)
To verify that this patch wouldn't cause much performance change,
the Borg tests were run. Here was the result:
       avg_psnr   overall_psnr  ssim
hdres: -0.002     0.006         0.013
midres:   0         0             0
lowres:   0         0             0

Change-Id: Iae395ae7b741e0513cf5bab9dcace110b792a67d

7 years agoMerge "Enhance the row mt sync read to accept the sync_range greater than 1"
Yunqing Wang [Mon, 3 Apr 2017 15:59:50 +0000 (15:59 +0000)]
Merge "Enhance the row mt sync read to accept the sync_range greater than 1"

7 years agoEnhance the row mt sync read to accept the sync_range greater than 1
Yunqing Wang [Tue, 28 Mar 2017 18:27:31 +0000 (11:27 -0700)]
Enhance the row mt sync read to accept the sync_range greater than 1

The row mt sync read uses sync_range = 1, and wouldn't work if we want
to use a sync_range that is greater than 1. To make it work, this sync
read code is modified. Pass in col instead of col - 1 to make it
consistent with other row mt code in VP9, and then add 1 in "while"
codition.

Change-Id: I4a0e487190ac5d47b8216368da12d80fec779c1a

7 years agovp9: SVC: Fix issue with artifact for svc-denoising.
Marco [Fri, 31 Mar 2017 00:08:44 +0000 (17:08 -0700)]
vp9: SVC: Fix issue with artifact for svc-denoising.

Issue/bug happens for denoising with spatial layers, where
the golden (spatial) reference is used in pickmode, but
denoising is only done wrt to last (temporal).

Fix is to make sure set_ref_ptrs is set before build predictors
in denoiser.

Change-Id: I793cf441341edf7c4a88b8ab1e1b22b3cb0eb508

7 years agoMerge changes from topic 'rm-dec-frame-parallel'
James Zern [Thu, 30 Mar 2017 19:07:43 +0000 (19:07 +0000)]
Merge changes from topic 'rm-dec-frame-parallel'

* changes:
  vpxdec: silently ignore -frame-parallel
  vp9: make VPX_CODEC_USE_FRAME_THREADING a no-op

7 years agovpxdec: silently ignore -frame-parallel
James Zern [Thu, 30 Mar 2017 06:31:53 +0000 (23:31 -0700)]
vpxdec: silently ignore -frame-parallel

BUG=webm:1395

Change-Id: Ibf47cc931e51b71e49067c6d7b7a39ab57c11c96

7 years agovp9: make VPX_CODEC_USE_FRAME_THREADING a no-op
James Zern [Thu, 30 Mar 2017 06:30:11 +0000 (23:30 -0700)]
vp9: make VPX_CODEC_USE_FRAME_THREADING a no-op

this is unmaintained and due to be removed

BUG=webm:1395

Change-Id: Iaffa6aa057c820fd1a182b93ebb45d4286e1306e

7 years agortcd,unit tests: fix ppc64 build
James Zern [Thu, 30 Mar 2017 05:51:22 +0000 (22:51 -0700)]
rtcd,unit tests: fix ppc64 build

match ppc* in rtcd to ensure vsx is included

Change-Id: I331a5d35e7160eeb69ebd14b98ba03ec5be6c600

7 years agovp9: SVC: fix to allow output of denoised result.
Marco [Wed, 29 Mar 2017 21:01:57 +0000 (14:01 -0700)]
vp9: SVC: fix to allow output of denoised result.

Change-Id: Iaf55cfb5e9621d074eb33d6a32f184e4777968f8

7 years agovp9: 1 pass SVC: Modify condition for intra-mode search.
Marco [Wed, 29 Mar 2017 16:13:56 +0000 (09:13 -0700)]
vp9: 1 pass SVC: Modify condition for intra-mode search.

Temporary override to condition for disallowing intra-search in SVC,
since golden (spatial) reference is currently suppressed due to
artifact issue.

Change-Id: I28ed7fdddc9fcdbcc0a4175a247a3ecc94c11767

7 years agoMerge "rate_hist: add parameter validation"
James Zern [Wed, 29 Mar 2017 02:41:07 +0000 (02:41 +0000)]
Merge "rate_hist: add parameter validation"