]> granicus.if.org Git - libvpx/log
libvpx
6 years agovp9-svc: Update layer_id of frame buffer idx last refreshed.
Marco Paniconi [Mon, 14 May 2018 05:17:17 +0000 (22:17 -0700)]
vp9-svc: Update layer_id of frame buffer idx last refreshed.

Remove some unused code and add parameter to keep track
of the layer_id of the frame buffer indices last refreshed.

This is useful for verifying constaints on spatial-temporal pattern,
for fixed/non-flexible mode.

Change-Id: I6957bb43157eb31df49dac1b8245facc043e4a49

6 years agoMerge "Fix valgrind failure on uninitialized values."
Jerome Jiang [Sat, 12 May 2018 06:08:16 +0000 (06:08 +0000)]
Merge "Fix valgrind failure on uninitialized values."

6 years agoFix valgrind failure on uninitialized values.
Jerome Jiang [Sat, 12 May 2018 00:15:10 +0000 (17:15 -0700)]
Fix valgrind failure on uninitialized values.

Change-Id: I917d884c9fab9b15bb092de5675f92225f1cdebd

6 years agovp9-svc: Fix pattern update for skip enhancement layers.
Marco Paniconi [Fri, 11 May 2018 22:49:34 +0000 (15:49 -0700)]
vp9-svc: Fix pattern update for skip enhancement layers.

Use the same logic as for dropped frames to be consistent.

Change-Id: I16fd317e70514fe8516d9eb350c275d1813e943e

6 years agoMerge "vp9-svc: Fix when whole superframe is dropped."
Marco Paniconi [Fri, 11 May 2018 17:45:11 +0000 (17:45 +0000)]
Merge "vp9-svc: Fix when whole superframe is dropped."

6 years agovp9-svc: Fix when whole superframe is dropped.
Marco Paniconi [Tue, 8 May 2018 05:54:48 +0000 (22:54 -0700)]
vp9-svc: Fix when whole superframe is dropped.

When the whole superframe is dropped (due to rate control),
don't increment the temporal layer counter.

This is a temporary fix to prevent an issue where temporal
prediction pattern is possibly broken.

Updated svc_datarate tests to handle this case.

Change-Id: Icac44fdc9d0f08a957776c937584db4b2c7927c7

6 years agoMerge "Faster VSX vpx_quantize_b"
Luc Trudeau [Fri, 11 May 2018 13:29:17 +0000 (13:29 +0000)]
Merge "Faster VSX vpx_quantize_b"

6 years agovpx_subtract_block_neon: add explicit cast
James Zern [Fri, 11 May 2018 06:08:54 +0000 (23:08 -0700)]
vpx_subtract_block_neon: add explicit cast

quiets ptrdiff_t -> int conversion warning

Change-Id: If6b545a736fc19e48e290961736b1618df97db3e

6 years agoMerge "Fix vpxdec fuzz failure."
Jerome Jiang [Fri, 11 May 2018 05:59:24 +0000 (05:59 +0000)]
Merge "Fix vpxdec fuzz failure."

6 years agoMerge "vp9: Adjust some early exits in nonrd-pickmode."
Marco Paniconi [Fri, 11 May 2018 04:45:13 +0000 (04:45 +0000)]
Merge "vp9: Adjust some early exits in nonrd-pickmode."

6 years agoRemove ppc64 (big endian) from configure
Luc Trudeau [Wed, 9 May 2018 19:08:53 +0000 (15:08 -0400)]
Remove ppc64 (big endian) from configure

Remove big endian PowerPC 64 from configure, as this build is problematic and
not supported. PowerPC 64 will be limited to little endian (ppc64le).

BUG=webm:1525
BUG=webm:1508

Change-Id: Id6a86d5913192549e03ac8f77879ba7526b752c8

6 years agoMerge "ppc: Add vpx_iwht4x4_16_add_vsx"
James Zern [Fri, 11 May 2018 03:17:55 +0000 (03:17 +0000)]
Merge "ppc: Add vpx_iwht4x4_16_add_vsx"

6 years agoMerge "configure: check for arm_neon.h w/neon builds"
James Zern [Fri, 11 May 2018 03:15:31 +0000 (03:15 +0000)]
Merge "configure: check for arm_neon.h w/neon builds"

6 years agoMerge "Update vpx_subtract_block_neon()"
James Zern [Fri, 11 May 2018 01:47:08 +0000 (01:47 +0000)]
Merge "Update vpx_subtract_block_neon()"

6 years agoFaster VSX vpx_quantize_b
Luc Trudeau [Thu, 10 May 2018 20:54:13 +0000 (16:54 -0400)]
Faster VSX vpx_quantize_b

Process 16 coefficients on the first iteration (a full 4x4) and 24 coefficients
on subsequent iteration.

VSX/VP9QuantizeTest.DISABLED_Speed
Before:
4x4   176 ms
8x8    91 ms
16x16  72 ms
After:
4x4   152 ms
8x8    82 ms
16x16  64 ms

Change-Id: I07cb130833504206ccdc5bc12ae5af369364999a

6 years agovp9: Adjust some early exits in nonrd-pickmode.
Marco Paniconi [Thu, 10 May 2018 21:27:33 +0000 (14:27 -0700)]
vp9: Adjust some early exits in nonrd-pickmode.

Condition some early exitis in nonrd-pickmode on the
motion vector, to make sure we always test (0, 0) for
inter-layer prediction.

Change-Id: Id0e790ecc75ccfb7031d3e8786ccdd13781d81fe

6 years agoUpdate vpx_subtract_block_neon()
Linfeng Zhang [Tue, 8 May 2018 23:37:07 +0000 (16:37 -0700)]
Update vpx_subtract_block_neon()

Change-Id: Ie2ac06c090c8f92268e9a799e96aa5192a1bdcd2

6 years agoMerge "Update vpx_comp_avg_pred_neon()"
James Zern [Thu, 10 May 2018 23:23:47 +0000 (23:23 +0000)]
Merge "Update vpx_comp_avg_pred_neon()"

6 years agoMerge "Remove extra line in warnings in ivfdec.c"
Jerome Jiang [Thu, 10 May 2018 23:19:26 +0000 (23:19 +0000)]
Merge "Remove extra line in warnings in ivfdec.c"

6 years agoconfigure: check for arm_neon.h w/neon builds
James Zern [Thu, 4 Aug 2016 18:54:02 +0000 (11:54 -0700)]
configure: check for arm_neon.h w/neon builds

fails at configure time rather than compile time unless using
--enable-external-build

Change-Id: I966ee1000e28fdcc3f4a29759789b056faee0010

6 years agoMerge "Make upper limit of frame size in ivf reader consistent."
Jerome Jiang [Thu, 10 May 2018 22:58:30 +0000 (22:58 +0000)]
Merge "Make upper limit of frame size in ivf reader consistent."

6 years agoFix vpxdec fuzz failure.
Jerome Jiang [Tue, 20 Mar 2018 17:34:26 +0000 (10:34 -0700)]
Fix vpxdec fuzz failure.

BUG=webm:1495

Change-Id: Ibaee35aa5e8e00847c61e707f2c9b4c0cff23673

6 years agoRemove extra line in warnings in ivfdec.c
Jerome Jiang [Thu, 10 May 2018 21:41:10 +0000 (14:41 -0700)]
Remove extra line in warnings in ivfdec.c

The warnings give an extra line which is confusing sometimes.

E.g.

Warning: Read invalid frame size (308164564) // This is for frame 5

Warning: Failed to decode frame 5: Invalid parameter
Warning: Read invalid frame size (1936229463) // This is for frame 6

Warning: Failed to decode frame 6: Invalid parameter
Warning: Read invalid frame size (2282536257)

Change-Id: I1753fa32079deca5c8b534c6ca9a527cc9e491e9

6 years agoMake upper limit of frame size in ivf reader consistent.
Jerome Jiang [Thu, 10 May 2018 21:30:54 +0000 (14:30 -0700)]
Make upper limit of frame size in ivf reader consistent.

Change the limit of frame size in ivf reader used by test to make it
consistent with ivf reader used in vpxdec.

Change-Id: I19ab05adf51eca65322e609efdf4d83ad66af847

6 years agoMerge "VSX version of vpx_quantize_b_vsx"
Luc Trudeau [Thu, 10 May 2018 13:48:07 +0000 (13:48 +0000)]
Merge "VSX version of vpx_quantize_b_vsx"

6 years agovp9-svc: Fix inter-layer early exit threshold.
Marco Paniconi [Thu, 10 May 2018 03:37:48 +0000 (20:37 -0700)]
vp9-svc: Fix inter-layer early exit threshold.

If the scale factors are 1 (no scaling), set the threshold
for skipping the inter-layer prediction to 0, so we will
more often test this mode.

Improves quality for upper layers for quality layers
in svc mode.

Change-Id: Iaf848d44f6cc153780db861b76517a4cf9672c45

6 years agoMerge "Don't use transform domain distortion when eob is 0"
Hui Su [Wed, 9 May 2018 22:09:11 +0000 (22:09 +0000)]
Merge "Don't use transform domain distortion when eob is 0"

6 years agoVSX version of vpx_quantize_b_vsx
Luc Trudeau [Wed, 9 May 2018 03:00:20 +0000 (23:00 -0400)]
VSX version of vpx_quantize_b_vsx

Low bit depth version only. Passes the VP9QuantizeTest.

Change-Id: I6546f872864bd404a7e353348b0554aab1de5bf0

6 years agoMerge "vp9-svc: Fix to SVC for frame dropping."
Marco Paniconi [Wed, 9 May 2018 17:41:33 +0000 (17:41 +0000)]
Merge "vp9-svc: Fix to SVC for frame dropping."

6 years agoMerge "Improved slide show coding"
Paul Wilkins [Wed, 9 May 2018 11:14:25 +0000 (11:14 +0000)]
Merge "Improved slide show coding"

6 years agoUpdate vpx_comp_avg_pred_neon()
Linfeng Zhang [Wed, 9 May 2018 00:37:18 +0000 (17:37 -0700)]
Update vpx_comp_avg_pred_neon()

Separate width 4 and 8 cases to reduce jumps in loop in clang.

Change-Id: I6ffc6f1555f2ad08b72a8dba35a78b9fd5f95a73

6 years agovp9-svc: Fix to SVC for frame dropping.
Marco Paniconi [Tue, 8 May 2018 18:05:03 +0000 (11:05 -0700)]
vp9-svc: Fix to SVC for frame dropping.

When the previous frame is dropped, for the current
spatial layer make sure the lst_fb_idx corresponds
to the buffer index last updated on the (last) encoded
TL0 frame(for same spatial layer).

This is needed to preserve the temporal prediction pattern
for fixed/non-flexible mode under frame dropping.

Change-Id: Ifc8e257beb025654a81580c4da0a181235724508

6 years agoUpdate SadMxNx4 NEON functions
Linfeng Zhang [Tue, 8 May 2018 19:52:10 +0000 (12:52 -0700)]
Update SadMxNx4 NEON functions

Change-Id: Ia313a6da00a05837fcd4de6ece31fa1c0016438c

6 years agoMerge "configure: Disable pthread_h if linking failed"
Martin Storsjö [Tue, 8 May 2018 19:04:12 +0000 (19:04 +0000)]
Merge "configure: Disable pthread_h if linking failed"

6 years agoMerge "Add vpx_sum_squares_2d_i16_neon()"
Linfeng Zhang [Tue, 8 May 2018 16:50:33 +0000 (16:50 +0000)]
Merge "Add vpx_sum_squares_2d_i16_neon()"

6 years agoImproved slide show coding
paulwilkins [Thu, 12 Apr 2018 16:14:25 +0000 (09:14 -0700)]
Improved slide show coding

This patch improves coding of slide shows with fade or other
complex transitions.

Previously, fades and other complex transitions between static "slides"
were sometimes being incorrectly marked such that they were coded
as a single static slide rather than two slides with a transition.

As the initial key frame for the first slide is not necessarily a good
predictor of the second slide and ARFs were turned off, this led to a
poor visual and metrics outcome in some such cases.

This patch allows for long GF groups in static sections before and after
a complex transition (instead of just with  simple slide transitions) with
one or more normal ARF groups during the transition. It also enforces a
single "normal" length GF group after the transition before any extended
group is allowed. The reason for this is that the ARF that spans the
transition my not have a very high quality and hence may not act as a
good GF for the long static section that follows.

Change-Id: Ica1f979e27d8a0625f3cebf7b7cf6d69edccaba9

6 years agoUpdate vpx_sum_squares_2d_i16_sse2()
Linfeng Zhang [Mon, 7 May 2018 20:38:04 +0000 (13:38 -0700)]
Update vpx_sum_squares_2d_i16_sse2()

Change-Id: I5a2ca2ed246277cf6b1ef2ffac34ce5c40aa0158

6 years agoAdd vpx_sum_squares_2d_i16_neon()
Linfeng Zhang [Mon, 7 May 2018 19:46:14 +0000 (12:46 -0700)]
Add vpx_sum_squares_2d_i16_neon()

Perf shows CPU time of this function dropped from 0.81% to 0.15%.

Change-Id: I8a7649ca5c15af2fc65cfb848f5befa0cc5e64f2

6 years agoDon't use transform domain distortion when eob is 0
Hui Su [Tue, 1 May 2018 04:11:31 +0000 (21:11 -0700)]
Don't use transform domain distortion when eob is 0

When eob is 0, pixel domain distortion is more accurate and efficient.

This mainly affects speed >= 2. Speed 0 always use pixel domain
distortion; speed 1 use it most of the time.

Compression impact(negative means gain):
        speed 2   speed 3   speed 4
lowres   -0.04%   -0.06%    -0.06%
midres   -0.10%   -0.10%    -0.20%
hdres    -0.01%   -0.03%    -0.06%

Encoding speed is about neutral.

Change-Id: I77b957658deeaad57381fd13afc11bacdec8c08f

6 years agoMerge "vp9-svc: Reset fb_idx for unused references."
Marco Paniconi [Fri, 4 May 2018 17:12:59 +0000 (17:12 +0000)]
Merge "vp9-svc: Reset fb_idx for unused references."

6 years agoconfigure: Disable pthread_h if linking failed
Martin Storsjo [Fri, 4 May 2018 06:32:41 +0000 (09:32 +0300)]
configure: Disable pthread_h if linking failed

When doing both check_header and check_lib, the check_header call
will already enable pthread_h if the header was found. This was
overlooked when the pthread linking check was amended into a header
check and a separate linking check in 9b7d4cce635e.

This brings back the same result as the original check in 38dc27cc6.

Change-Id: I0efb38f5780f7c79e2eb2b14290d6094096ea222

6 years agovp9-svc: Add memset on the svc fb_idx.
Marco Paniconi [Fri, 4 May 2018 02:42:08 +0000 (19:42 -0700)]
vp9-svc: Add memset on the svc fb_idx.

The memset is added to better handle frame drops
with the GET_SVC_REF_FRAME_CONFIG contro

There is an issue with some tests in bypass mode,
so condition it on that for now.

Change-Id: I2635037143f14ff62a36be7c22b2b604a0c1efc2

6 years agovp9-svc: Reset fb_idx for unused references.
Marco Paniconi [Mon, 30 Apr 2018 20:42:26 +0000 (13:42 -0700)]
vp9-svc: Reset fb_idx for unused references.

For fixed (non-flexible) SVC mode.
No change in behavior.

Needed for future change to make Intra-only frame work.

Change-Id: I91e18776e7ef27c9c6fcbc8d5f64764d9cc3d9a9

6 years agovp9-svc: On key frame update all reference slots for SVC.
Marco Paniconi [Thu, 3 May 2018 01:04:35 +0000 (18:04 -0700)]
vp9-svc: On key frame update all reference slots for SVC.

Key frame updates the slots corresponding to the 3 references
last/golden/altref, but for SVC where more references buffers
may be in use, especialy for dynamically swithing up/down in layers,
make sure we should update all 8 slots on key frame.

Change-Id: Ifcca12608f420d5bae32b92794a3afe9b6369f77

6 years agoMerge "Clean switch cases in vp9 encoder"
Linfeng Zhang [Wed, 2 May 2018 20:01:59 +0000 (20:01 +0000)]
Merge "Clean switch cases in vp9 encoder"

6 years agoMerge "vp9: [loongson] optimize vpx_convolve8 with mmi"
xiwei gu [Wed, 2 May 2018 08:54:24 +0000 (08:54 +0000)]
Merge "vp9: [loongson] optimize vpx_convolve8 with mmi"

6 years agoClean switch cases in vp9 encoder
Linfeng Zhang [Wed, 2 May 2018 00:57:51 +0000 (17:57 -0700)]
Clean switch cases in vp9 encoder

To save a branch.

Change-Id: Ifa2be7583e95c6991784731c654bbd4cce31e993

6 years agoMerge "Update variance_test.cc"
James Zern [Sat, 28 Apr 2018 05:52:45 +0000 (05:52 +0000)]
Merge "Update variance_test.cc"

6 years agoUpdate variance_test.cc
Linfeng Zhang [Wed, 25 Apr 2018 23:04:45 +0000 (16:04 -0700)]
Update variance_test.cc

Change-Id: I1301781f0f2528a61ad2b5c2828404b2b3e3e8b9

6 years agovp9-svc: Remove the memset on the svc fb_idx.
Marco Paniconi [Fri, 27 Apr 2018 22:35:31 +0000 (15:35 -0700)]
vp9-svc: Remove the memset on the svc fb_idx.

This fixes failures on the datarate tests for
temporal layers with frame dropping.

The memset was only added to better handle frame drops
with the GET_SVC_REF_FRAME_CONFIG control from 43c58df3.
So ok to remove it for now.

Change-Id: I256d9ac4278b93fe6f39b94cce2e458a1a5eff69

6 years agoMerge "VP9 SVC: Add new level to constrain inter-layer pred."
Marco Paniconi [Fri, 27 Apr 2018 04:55:22 +0000 (04:55 +0000)]
Merge "VP9 SVC: Add new level to constrain inter-layer pred."

6 years agoMerge "Respect MV limit in vp9_int_pro_motion_estimation()"
Hui Su [Fri, 27 Apr 2018 04:41:52 +0000 (04:41 +0000)]
Merge "Respect MV limit in vp9_int_pro_motion_estimation()"

6 years agoVP9 SVC: Add new level to constrain inter-layer pred.
Jerome Jiang [Wed, 25 Apr 2018 18:33:24 +0000 (11:33 -0700)]
VP9 SVC: Add new level to constrain inter-layer pred.

Add another level (INTER_LAYER_PRED_ON_CONSTRAINED) to the
inter-layer prediction control. This new level enforces the
condition that a given spatial layer S can only do inter-layer
prediction from the previous spatial layer (S - 1) from the same
time/superframe.

BUG=webm:1526

Change-Id: I0a1ec95b2c220c7b13a9a425d5fb0a8814c23c70

6 years agovp9-svc: Remove unneeded call and init some parameters.
Marco Paniconi [Fri, 27 Apr 2018 00:32:22 +0000 (17:32 -0700)]
vp9-svc: Remove unneeded call and init some parameters.

Remove the unneeded vp9_copy_flags_ref_update_idx(cpi),
and initialize the struct parameters needed for the
GET_SVC_REF_FRAME_CONFIG. This init is useful for the case
for spatial layer frame drops.

Change-Id: If89e8349f6246c33720ecbb758d41a932d21e496

6 years agoMerge "Do one level less of transform search for large blocks"
Hui Su [Thu, 26 Apr 2018 23:58:33 +0000 (23:58 +0000)]
Merge "Do one level less of transform search for large blocks"

6 years agoRespect MV limit in vp9_int_pro_motion_estimation()
Hui Su [Thu, 26 Apr 2018 20:31:43 +0000 (13:31 -0700)]
Respect MV limit in vp9_int_pro_motion_estimation()

Change-Id: I08cb072a32e06c6452eca068b2f7ef7287f221e6

6 years agoDo one level less of transform search for large blocks
Hui Su [Wed, 25 Apr 2018 05:42:54 +0000 (22:42 -0700)]
Do one level less of transform search for large blocks

If block size is larger than 32x32, search transform size for one level
less than the other blocks.

This mainly affects speed 0 and 1, as speed >= 2 uses largest transform
size(except for keyframes and alt-ref frames).

Compression(negative means gain):
        speed 0     speed 1
lowres  -0.007%       0.00%
midres   0.023%     -0.011%
hdres    0.002%     -0.016%

Encoder speed:
Tested on crowd_run_1080p 30 frames
Fixed QP = 30, speed 0: 582.5s -> 564.6s
               speed 1:  75.0s ->  73.3s

Change-Id: I46622efafe0e88d502efa1480a5324ead1d1e8d0

6 years agoMerge "Calculate transform size cost once per frame"
Hui Su [Wed, 25 Apr 2018 16:25:58 +0000 (16:25 +0000)]
Merge "Calculate transform size cost once per frame"

6 years agoMerge "Add speed feture to control tx size search depth"
Hui Su [Wed, 25 Apr 2018 16:25:40 +0000 (16:25 +0000)]
Merge "Add speed feture to control tx size search depth"

6 years agoMerge "vp9-svc: Add GET control to get SVC pattern info."
Jerome Jiang [Wed, 25 Apr 2018 04:59:10 +0000 (04:59 +0000)]
Merge "vp9-svc: Add GET control to get SVC pattern info."

6 years agovp9-svc: Add GET control to get SVC pattern info.
Marco Paniconi [Tue, 24 Apr 2018 04:03:47 +0000 (21:03 -0700)]
vp9-svc: Add GET control to get SVC pattern info.

Copy ref frame index in SVC struct after set in encoder.
Rename ext_{lst,gld,alt}_fb_idx to {lst,gld,alt}_fb_idx.

Bump up ABI version.

BUG=webm:1527

Change-Id: I06209040cb83d374030f40b79f0b36b0efe9f97d

6 years agovp9: [loongson] optimize vpx_convolve8 with mmi
guxiwei-hf@loongson.cn [Wed, 25 Apr 2018 01:55:05 +0000 (09:55 +0800)]
vp9: [loongson] optimize vpx_convolve8 with mmi

1. vpx_convolve_avg_mmi
2. vpx_convolve8_avg_horiz_mmi

Change-Id: Ie544aac45b4b1c0a0e51b44b650189ae5e88aee1

6 years agoFix vp8_multi_resolution_encoder test failure.
Jerome Jiang [Tue, 24 Apr 2018 23:37:18 +0000 (16:37 -0700)]
Fix vp8_multi_resolution_encoder test failure.

BUG=webm:1528

Change-Id: I8eb8278c2e61577159308dd5329be0577b82d1a6

6 years agoCalculate transform size cost once per frame
Hui Su [Tue, 24 Apr 2018 18:49:22 +0000 (11:49 -0700)]
Calculate transform size cost once per frame

Instead of doing it in every transform search loop.

Change-Id: I12dc402a6633d1a27d32cb6b58710b8c0ebf0fd4

6 years agoRemove get_tx_probs2()
Hui Su [Tue, 24 Apr 2018 00:15:23 +0000 (17:15 -0700)]
Remove get_tx_probs2()

This function is redundant.

Change-Id: I7651fc34787c09e59cb1366495f6b525dec8510d

6 years agoAdd speed feture to control tx size search depth
Hui Su [Mon, 23 Apr 2018 21:05:23 +0000 (14:05 -0700)]
Add speed feture to control tx size search depth

Set the max depth as 2 for speed 0.

Compression(negative means gain):
        speed 0     speed 1
lowres   -0.01%      0.00%
midres    0.05%     -0.01%
hdres    -0.01%      0.01%

Encoding speed gain:
Tested on crowd_run_1080p 30 frames
Fixed QP = 20, speed 0: 669.7s -> 656.1s
               speed 1: 104.5s -> 101.5s
Fixed QP = 40, speed 0: 440.7s -> 435.8s
               speed 1:  47.7s ->  45.1s

Change-Id: I61bc13818c72317b9f1d596727d54a906b20c012

6 years agoMerge changes I0202a556,Iebb98f3b
James Zern [Thu, 19 Apr 2018 21:29:42 +0000 (21:29 +0000)]
Merge changes I0202a556,Iebb98f3b

* changes:
  Update variance avx2 functions
  Update variance sse2 functions

6 years agoconfigure: Use both check_header and check_lib for pthreads
Martin Storsjo [Thu, 19 Apr 2018 05:55:44 +0000 (08:55 +0300)]
configure: Use both check_header and check_lib for pthreads

check_lib can be a stub that always returns true - make sure to
still use check_headers as before 38dc27cc6.

Change-Id: I5d471de56b16c015a0b686fa6c6caefa35bb89b4

6 years agovp9:aq-mode=3: Keep perc_refresh fixed for screen content mode.
Marco Paniconi [Thu, 19 Apr 2018 02:59:45 +0000 (19:59 -0700)]
vp9:aq-mode=3: Keep perc_refresh fixed for screen content mode.

Don't allow for changing the perc_refresh with screen-content
mode, as this helps reduce some overshoot for static content.

Change-Id: Idbe1849e7a14ef18fda20bee6dced809f134b7f7

6 years agoMerge "vp9: Rate control fix for CBR mode."
Marco Paniconi [Thu, 19 Apr 2018 02:56:26 +0000 (02:56 +0000)]
Merge "vp9: Rate control fix for CBR mode."

6 years agoMerge "vp9: Remove limit on QP on key frame for CBR."
Marco Paniconi [Thu, 19 Apr 2018 01:39:52 +0000 (01:39 +0000)]
Merge "vp9: Remove limit on QP on key frame for CBR."

6 years agovp9: Add condition of real-time mode to scene detection.
Marco Paniconi [Thu, 19 Apr 2018 00:46:16 +0000 (17:46 -0700)]
vp9: Add condition of real-time mode to scene detection.

This was removed by error from the change: ce11afb, and
made some datarate tests fail.

Change-Id: I0c29e1f5aede8f56ce835b25fed0528722350241

6 years agovp9: Rate control fix for CBR mode.
Marco Paniconi [Thu, 19 Apr 2018 00:15:16 +0000 (17:15 -0700)]
vp9: Rate control fix for CBR mode.

For CBR mode: modify the qp clamping to allow q to respond
faster to overshoot. Can reduce some suprious overshoot events
observed in screen content coding.

Change-Id: I0b3f54b0d1b4086182f834e557a4121950b176d4

6 years agoMerge "SVC: Fix duplicated run of svc datarate tests."
Jerome Jiang [Wed, 18 Apr 2018 23:15:57 +0000 (23:15 +0000)]
Merge "SVC: Fix duplicated run of svc datarate tests."

6 years agovp9: Remove limit on QP on key frame for CBR.
Marco Paniconi [Wed, 18 Apr 2018 22:54:03 +0000 (15:54 -0700)]
vp9: Remove limit on QP on key frame for CBR.

This piece was carried over the VBR routine, for CBR
mode we don't want to apply this limit.

Change-Id: Ib9e9937eabeff8cfd30e11c9bd17444cc2b591aa

6 years agoSVC: Fix duplicated run of svc datarate tests.
Jerome Jiang [Mon, 16 Apr 2018 23:37:04 +0000 (16:37 -0700)]
SVC: Fix duplicated run of svc datarate tests.

Change-Id: I3f4e45b398009852f1183943461625d621c4eb80

6 years agovp9: Changes for scene detection overshoot and SVC.
Marco Paniconi [Mon, 16 Apr 2018 23:15:05 +0000 (16:15 -0700)]
vp9: Changes for scene detection overshoot and SVC.

Refactor the scene detection for 1 pass cbr to allow the
scene detection to be checked once per superframe (on the base layer),
using the full resolution sources.

If scene change is detected: check for re-encoding due to
large overshoot for all spatial layers withing the superframe.

Add speed feature to control the re-encode step.
Keep the re-encode step on for now.

Small change in nonrd_pickmode to remove the possible skip of golden
reference for SVC, when the high_source_sad is set for the superframe.

Change only affects SVC encoding with screen-content mode enabled.

Change-Id: If4cfb52cb0dd0f0fce1c4214fa8b413f8f803d56

6 years agoMerge "configure: Test linking pthreads before using it"
Martin Storsjö [Wed, 18 Apr 2018 06:23:09 +0000 (06:23 +0000)]
Merge "configure: Test linking pthreads before using it"

6 years agoMerge "Clean up svc comment in vp9_bitstream.c"
Jerome Jiang [Tue, 17 Apr 2018 23:30:23 +0000 (23:30 +0000)]
Merge "Clean up svc comment in vp9_bitstream.c"

6 years agoUpdate variance avx2 functions
Linfeng Zhang [Tue, 17 Apr 2018 20:43:16 +0000 (13:43 -0700)]
Update variance avx2 functions

Old vs New
Variance 64x64 time:  1145 ms           797 ms
Variance 64x32 time:  1200 ms           831 ms
Variance 32x32 time:  1228 ms          1135 ms
Variance 32x16 time:  1374 ms          1491 ms
Variance 16x16 time:  1688 ms          1571 ms

sse2 vs avx2
Variance 32x64 time:  1645 ms           957 ms
Variance 16x32 time:  2031 ms          1243 ms
Variance 16x8  time:  3071 ms          2275 ms

Change-Id: I0202a556e4629977d647e219c2e897e1ab6accb2

6 years agovp9: Remove this_key_frame_forced setting for CBR.
Marco Paniconi [Tue, 17 Apr 2018 17:32:25 +0000 (10:32 -0700)]
vp9: Remove this_key_frame_forced setting for CBR.

The setting this_key_frame_forced can lead to large key frame sizes,
not suitable for CBR rate control used for RTC.

Change-Id: Idf6d2bf385d5b1494f4bf783f623b7c202f34e55

6 years agoUpdate variance sse2 functions
Linfeng Zhang [Mon, 16 Apr 2018 21:44:00 +0000 (14:44 -0700)]
Update variance sse2 functions

Old vs New
Variance 64x64 time:   197 ms   143 ms
Variance 64x32 time:   200 ms   146 ms
Variance 32x64 time:   203 ms   140 ms
Variance 32x32 time:   214 ms   152 ms
Variance 32x16 time:   243 ms   153 ms
Variance 16x32 time:   234 ms   197 ms
Variance 16x16 time:   205 ms   205 ms
Variance 16x8  time:   228 ms   222 ms
Variance 8x16  time:   228 ms   232 ms
Variance 8x8   time:   282 ms   240 ms
Variance 8x4   time:   506 ms   341 ms
Variance 4x8   time:   518 ms   415 ms
Variance 4x4   time:   604 ms   628 ms

Observed vp9 encoder speed up when encoding a 720p video.

Change-Id: Iebb98f3b3d8adbc11a733a529d8427ce3d2a5314

6 years agovp9: fix multiple run for vp9 datarate tests using one bitrate.
Jerome Jiang [Mon, 16 Apr 2018 21:25:51 +0000 (14:25 -0700)]
vp9: fix multiple run for vp9 datarate tests using one bitrate.

Change-Id: I1458bf25fadc23e8be5a9532a153d7129a53accf

6 years agoMerge "vp9: refactor vp9 datarate test for better sharding."
Jerome Jiang [Sun, 15 Apr 2018 07:03:16 +0000 (07:03 +0000)]
Merge "vp9: refactor vp9 datarate test for better sharding."

6 years agoconfigure: Test linking pthreads before using it
Martin Storsjo [Sat, 14 Apr 2018 20:40:46 +0000 (23:40 +0300)]
configure: Test linking pthreads before using it

This avoids enabling pthreads if only pthreads-w32 is available.
pthreads-w32 provides pthread.h but has a link library with a
different name (libpthreadGC2.a).

Generally, always using win32 threads when on windows would be
sensible.

However, libstdc++ can be configured to use pthreads (winpthreads), and
in these cases, standard C++ headers can pollute the namespace with
pthreads declarations, which break the win32 threads headers that
declare similar symbols - leading us to prefer pthreads on windows
whenever available (see d167a1ae and bug 1132).

Change-Id: Icd668ccdaf3aeabb7fa4e713e040ef3d67546f00

6 years agovp9: refactor vp9 datarate test for better sharding.
Jerome Jiang [Thu, 12 Apr 2018 21:37:56 +0000 (14:37 -0700)]
vp9: refactor vp9 datarate test for better sharding.

Change-Id: Icfaf29e1ca847ba9e3748700c9e09383ce8d1f65

6 years agoClean up is_two_pass_svc.
Jerome Jiang [Fri, 13 Apr 2018 22:36:39 +0000 (15:36 -0700)]
Clean up is_two_pass_svc.

Change-Id: I9e92616471be380d3ba4e2b85399d7eb9f687d2f

6 years agoClean up svc comment in vp9_bitstream.c
Jerome Jiang [Fri, 13 Apr 2018 20:56:45 +0000 (13:56 -0700)]
Clean up svc comment in vp9_bitstream.c

SVC related conditions were removed long time ago and comments should be
removed too.

Change-Id: Iff3f3b6815d85ae5a69994932a4893cd1f831ce3

6 years agoMerge "vp9 svc: Refactor svc datarate test for better sharding."
Jerome Jiang [Fri, 13 Apr 2018 03:40:50 +0000 (03:40 +0000)]
Merge "vp9 svc: Refactor svc datarate test for better sharding."

6 years agoMerge "Silence warning when built with --enable-internal-stats."
Jerome Jiang [Fri, 13 Apr 2018 03:40:20 +0000 (03:40 +0000)]
Merge "Silence warning when built with --enable-internal-stats."

6 years agoSilence warning when built with --enable-internal-stats.
Jerome Jiang [Thu, 12 Apr 2018 22:05:58 +0000 (15:05 -0700)]
Silence warning when built with --enable-internal-stats.

Change-Id: I3a600a9baf2b8e46c109f4ec2b5bd6bafda4bf58

6 years agovp9 svc: Refactor svc datarate test for better sharding.
Jerome Jiang [Wed, 11 Apr 2018 23:32:00 +0000 (16:32 -0700)]
vp9 svc: Refactor svc datarate test for better sharding.

Wrap denoiser tests under config flags.

Change-Id: I6175c3c9d8b5b079ad35a55553383145db58a10f

6 years agoMerge "Add extra case to wq_err_divisor()"
Paul Wilkins [Thu, 12 Apr 2018 10:42:50 +0000 (10:42 +0000)]
Merge "Add extra case to wq_err_divisor()"

6 years agovpx_image: remove unused image formats
James Zern [Mon, 9 Apr 2018 23:07:40 +0000 (16:07 -0700)]
vpx_image: remove unused image formats

libvpx only emits:
VPX_IMG_FMT_{I420,I422,I440,I444,I42016,I42216,I44016,I44416}
and additionally supports YV12 as input.

interleaved yuv, rgb and alpha formats are unused.

Change-Id: Ie2ab1099e950c6e696f475d46882f5c47a174042

6 years agoMerge "vp9-svc: Make constrained_layer_drop default for svc."
Marco Paniconi [Mon, 9 Apr 2018 18:05:51 +0000 (18:05 +0000)]
Merge "vp9-svc: Make constrained_layer_drop default for svc."

6 years agoMerge "Fix settings for num of tiles in samples & tests."
Jerome Jiang [Mon, 9 Apr 2018 17:03:59 +0000 (17:03 +0000)]
Merge "Fix settings for num of tiles in samples & tests."

6 years agovp9-svc: Make constrained_layer_drop default for svc.
Marco Paniconi [Mon, 9 Apr 2018 16:41:47 +0000 (09:41 -0700)]
vp9-svc: Make constrained_layer_drop default for svc.

Switch the order of constrained and layer drop mode,
and keep constrained_layer_drop as the default.
Update the svc datarate tests.

Change-Id: I764270f7b4964b87b0cd3da6c2f96a628f212a30

6 years agoFix settings for num of tiles in samples & tests.
Jerome Jiang [Thu, 5 Apr 2018 21:45:06 +0000 (14:45 -0700)]
Fix settings for num of tiles in samples & tests.

The control is set by log2 of number of threads (such that the number of
tiles is the same of number of threads).

Thus it should be log2(num_threads) instead of (num_threads >> 1).

Change-Id: I2ccec5557e660048dad3e561534e1c74fc8eec1f

6 years agovp9-svc: Hybrid search on spatial layers whose base is key.
Marco Paniconi [Fri, 6 Apr 2018 16:17:46 +0000 (09:17 -0700)]
vp9-svc: Hybrid search on spatial layers whose base is key.

For spatial layers whose base is a key frame, i.e., when
svc.layer_context[cpi->svc.temporal_layer_id].is_key_frame = 1,
allow for hybrid search, similar to what we do on key frames.

For small blocks (<= 8x8) rd-based intra search will be used,
otherwise non-rd pick mode is used.

Feature is controlled by nonrd_keyframe, which is set to 1
for now on non-base spatial layers, so this change has
currently no effect.

Small change only when inter-layer prediction is off, as we now
call vp9_pick_intra_mode instead of vp9_pick_inter_mode on key frame.
But this change is very small/insignificant.

Change-Id: I5372470f720812926ebbe6c4ce68c04336ce0bdd

6 years agoMerge "remove support for yuv 411"
Johann Koenig [Fri, 6 Apr 2018 15:26:59 +0000 (15:26 +0000)]
Merge "remove support for yuv 411"