Commit Graph

11655 Commits

Author SHA1 Message Date
Dmitry Kovalev
70092af5c0 Cleaning up and speeding up vp9_idct32x32_1024_add_sse2().
Change-Id: If91017b792572c9db6e257011ca307bef8428486
2014-09-05 18:12:30 -07:00
Dmitry Kovalev
89963bf586 Merge "Removing postproc mmx code." 2014-09-05 18:11:08 -07:00
Yunqing Wang
1092140379 No longer use use_lastframe_partitioning speed feature
The speedup in rd_pick_partition() function makes it possible
to drop use_lastframe_partitioning feature. By doing that, we
achieve good PSNR gain with small speed loss. Also, this makes
encoding loop less complicated. The code cleanup patch will
follow.

Borg tests showed:
1. At speed 2,
   stdhd set: 0.201% PSNR gain, 0.133% SSIM gain;
   derf set:  0.262% PSNR gain, 0.276% SSIM gain.
2. At speed 3,
   stdhd set: 0.139% PSNR gain, 0.109% SSIM gain;
   derf set:  0.447% PSNR gain, 0.442% SSIM gain.

The average speed loss over selected test clips is within 1%
with the worst case of 4%.

Change-Id: Icfd2ded7869372b585a6972855d933b3d0280d90
2014-09-05 16:24:41 -07:00
Yunqing Wang
ebac8f3487 Merge "Correct the mode decisions in special cases" 2014-09-05 13:45:41 -07:00
Minghai Shang
a8d44b9915 [spatial svc]Add layer bitrates options and clean up parsing options from string
Change-Id: I6e8d3f08c8a8ce3555a498feba471ba9fec411c6
2014-09-05 12:44:17 -07:00
Dmitry Kovalev
54bec0971f Merge "Initializing intra modes without vpx_once()." 2014-09-05 12:03:36 -07:00
James Zern
46faaeeffb Merge changes I7b9f40dc,I76e74f2e
* changes:
  vp9: correct context buffer resize check
  vp9: fail decode if block/frame refs are corrupt
2014-09-05 12:01:59 -07:00
Yunqing Wang
1dd9a63929 Correct the mode decisions in special cases
The rate costs calculated for inter modes are not precise in some
cases, which causes NEWMV is chosen instead of NEARESTMV, NEARMV,
and ZEROMV. This patch added checks for these cases, and corrected
the mode decisions.

Borg tests at speed 3 showed:
1. stdhd set: 0.102% PSNR gain and 0.088% SSIM gain.
2. derf set:  0.147% PSNR gain and 0.132% SSIM gain.
No speed change.

Change-Id: I35d17684b89ad4734fb610942d707899146426db
2014-09-05 12:01:07 -07:00
James Zern
6f980c6a1e Merge "fix x86-darwin* build" 2014-09-05 11:58:55 -07:00
James Zern
2886b91790 Merge "vp9: skip loopfilter when the frame is corrupt" 2014-09-05 11:58:09 -07:00
Dmitry Kovalev
1100e262c5 Removing postproc mmx code.
Removed functions:
* vp9_post_proc_down_and_across_mmx
* vp9_mbpost_proc_down_mmx
* vp9_plane_add_noise_mmx

They all have sse2 equivalent.

Change-Id: I59c1fac12b7c96ca4538d455e4400c2b7875feff
2014-09-05 11:52:50 -07:00
Dmitry Kovalev
02a0c51e50 Merge "Adding temp cpi var." 2014-09-05 10:31:41 -07:00
Jia Jia
c97f5e8b86 vp8 common: change 'HAVE_NEON_ASM' to 'HAVE_NEON' for compiling functions of NEON intrinsics.
Change-Id: I975e5eac16f8b623ff589f0ec072cdaff2183b04
2014-09-05 12:24:05 +00:00
James Zern
a8083449e9 fix x86-darwin* build
vp9_variance_sse2.c contains a mix of intrinsics and references to
assembly which uses x86inc.asm; it's conditionally included as a result.

Change-Id: I254451483a65881c0b8e18e27bf0c3ddef60c4ec
2014-09-04 23:32:13 -07:00
James Zern
35fadf1d25 bilinearpredict_neon: fix type conversion warnings
make bifilter4_coeff[][] uint8_t, no values exceed this range and
they're loaded with vdup_n_u8().

Change-Id: I921983e9edd828d29820e40ac30a7801dbe0fb4f
2014-09-04 20:50:42 -07:00
James Zern
bb4950dfdf vp9: correct context buffer resize check
allocations within vp9_alloc_context_buffers() rely on mi_rows/mi_cols
individually, use those to determine whether to realloc rather than
stride and stride * rows. this fixes a crash with some fuzzed files for
invalid accesses into last_frame_seg_map and above_context.

Change-Id: I7b9f40dcf170d443890f3bd2acd285507943c7d4
2014-09-04 19:14:21 -07:00
James Zern
440f5097c7 vp9: fail decode if block/frame refs are corrupt
proceeding using a corrupt (incompletely decoded) frame reference may
lead to incorrect assumptions about allocation sizes leading to a crash.

Change-Id: I76e74f2e1be127c2e2c7e1174bb3307497dfd23d
2014-09-04 19:14:00 -07:00
Dmitry Kovalev
c7b925c3fe Merge "Removing sz member from vpx_codec_priv. " 2014-09-04 17:28:22 -07:00
Dmitry Kovalev
ce1c9228d4 Merge "Removing unused function prototypes." 2014-09-04 17:28:16 -07:00
Yaowu Xu
dbdb87b72f Fix a visual studio warning
Change-Id: I7f935e61c8c0ade7893a5212701f07bb346d78cb
2014-09-04 17:00:32 -07:00
JackyChen
f8e5105b47 Merge "Map motion magnitude in VP9 denoiser." 2014-09-04 16:59:53 -07:00
Jingning Han
d435148fe6 Enable adaptive motion search for ARF coding
This commit turns on adaptive motion search for ARF coding, in
addition to other normal inter frame coding. It improves the
average compression efficiency:

stdhd 0.1%
derf  0.04%

For the test sequences, the speed 3 runtime is reduced:

pedestrian 1080p 2000 kbps, 149932 ms -> 144580 ms, (3.3% speed-up)
bus CIF 1000 kbps, 8050 ms -> 7895 ms, (1.9%)
highway CIF 100 bkps, 45033 ms -> 44078 ms, (2.2%)

Change-Id: I5228565b609f99e8ae04f6140a2bf2b64a831d21
2014-09-04 16:26:40 -07:00
Jingning Han
3de038f396 Merge "Speed up compound inter prediction mode check" 2014-09-04 16:09:07 -07:00
JackyChen
b869b970c1 Merge "Update the condition when COPY_BLOCK is chosen." 2014-09-04 15:48:22 -07:00
Dmitry Kovalev
46b83391e2 Merge "Removing local set_speed_features() function." 2014-09-04 15:36:52 -07:00
JackyChen
b1153f34d4 Map motion magnitude in VP9 denoiser.
This is to keep the same with VP8 denoiser.
If motion magnitude is small,
make denoiser more aggressive.

Change-Id: I942a6e2f2ed9aec6f0c4c1f9e5fa47066cadcc0c
2014-09-04 14:53:33 -07:00
Dmitry Kovalev
7897059e8b Adding temp cpi var.
Change-Id: Ifa3c1cc2317c1bc21d1042b9662b35056d1e9ed0
2014-09-04 14:51:29 -07:00
Dmitry Kovalev
91998e638e Removing sz member from vpx_codec_priv.
Change-Id: I811526a9ee9f237604f72abe7fc677e39e0f457f
2014-09-04 14:47:42 -07:00
JackyChen
d75266f141 Update the condition when COPY_BLOCK is chosen.
The change is just to keep the condition the same with VP8.

Change-Id: I9662b40996126605945dd853c0cbe8916c1ce578
2014-09-04 14:28:12 -07:00
Dmitry Kovalev
0f0382af46 Merge "Actually resetting random generator for all variance test cases." 2014-09-04 14:27:04 -07:00
Dmitry Kovalev
490943552f Removing unused function prototypes.
Change-Id: Ia5e383e2cf18052f6f1eacf8b9495ab8e4d58878
2014-09-04 14:26:30 -07:00
JackyChen
7ba600dc89 Merge "Fix a bug in VP9 denoiser." 2014-09-04 14:16:26 -07:00
Dmitry Kovalev
27db51c602 Merge "Adding sse2 variant for vp9_mse{8x8, 8x16, 16x8}." 2014-09-04 13:59:37 -07:00
Dmitry Kovalev
202edb3d23 Actually resetting random generator for all variance test cases.
Calling Reset(int) method instead of overloaded operator()(int).
Adding underscore at the end of class member name.

Change-Id: I01934e7bc056d4b594e5d05d693328febd34ac3c
2014-09-04 12:24:52 -07:00
JackyChen
e30f7698f5 Fix a bug in VP9 denoiser.
When the first try of denoising turns out to be too much,
we will use a softer filter by adopting an adjustment to
make the result closer to original pixel (as in VP8 denoiser).
The old code made the adjustment in the wrong direction.

Change-Id: I84e28fa9e01eef47c5a37d5a2e6d3d378a06786b
2014-09-04 11:46:36 -07:00
James Zern
f61e00c79d Merge "arm: Fix building vp8_subpixelvariance_neon.c with MSVC" 2014-09-04 11:00:53 -07:00
Scott LaVarnway
ec94967ffe Revert "Revert "VP8 for ARMv8 by using NEON intrinsics 10""
This reverts commit 677fb5123e

Compiles with 4.6.

Change-Id: I7f87048911b6bc28a61741d95501fa45ee97b819
2014-09-04 08:51:20 -07:00
Jia Jia
a51704d9c7 vp8 common: change 'HAVE_NEON_ASM' to 'HAVE_NEON' for compiling idct_blk_neon.c.
Change-Id: Ib89107fb824b5fe58afef6841104d5a27b2e0f2d
2014-09-04 08:40:26 -07:00
Martin Storsjo
0002da32e6 arm: Fix building vp8_subpixelvariance_neon.c with MSVC
Use the right return values - vget_low_s64 returns int64x1_t, not
a normal int64_t.

Also make __builtin_prefetch a no-op on MSVC for this file.

Change-Id: I4d2fce01d0ba106b98d3d53b137803119c2c2c08
2014-09-04 09:49:30 +03:00
Dmitry Kovalev
3820f568da Merge "Consistent allocation of vpx_codec_alg_priv_t." 2014-09-03 19:41:28 -07:00
Dmitry Kovalev
48197f0a70 Adding sse2 variant for vp9_mse{8x8, 8x16, 16x8}.
Change-Id: I6786d25ce4f32b8d8912f2d239a45ca15b310c4b
2014-09-03 19:02:14 -07:00
Dmitry Kovalev
ab73dba65f Merge "Replacing asm 16x16 variance calculation with intrinsics." 2014-09-03 18:57:33 -07:00
Dmitry Kovalev
406404af63 Merge "Small cleanup: reusing existing code." 2014-09-03 18:57:25 -07:00
Yaowu Xu
c9448ade2b Merge "Add a more complex SEG_LVL_SKIP test vector." 2014-09-03 17:17:12 -07:00
Jingning Han
d62d804e64 Speed up compound inter prediction mode check
This commit allows the encoder to store outcomes of single reference
frame modes and compares them to decide if the inter prediction
filter, forward transform, and quantization can be skipped.

The compression performance of speed 3 is down
derf  -0.364%
stdhd -0.198%

For test sequences, the speed 3 runtime is reduced
highway CIF 100 kbps, 51976 ms -> 45033 ms, 13% speed-up
stockholm 720p 1000 kbps, 71826 ms -> 67838 ms, 5.5% speed-up
pedestrian 1080p 2000 kbps, 154924 ms -> 150702 ms, 2.6% speed-up

Change-Id: I5aa26f918d2b4b5197a2c0afa2779319f1c88e44
2014-09-03 15:28:01 -07:00
Alex Converse
81e5d3f988 Add a more complex SEG_LVL_SKIP test vector.
This file changes the segmentation map, deactivates the map, and
reactivates the map.

Change-Id: I1c8b8949887257b9f4c49a1a9ba9a7c74666ef2e
2014-09-03 15:08:49 -07:00
Yaowu Xu
7ab5de04fd Merge "Change last_partition_redo_frequency for speed 3" 2014-09-03 14:57:02 -07:00
Yaowu Xu
44879ceea7 Merge "Remove redundant code" 2014-09-03 14:55:28 -07:00
Marco
01d9023de0 Merge "Make process_denoise_mode_change function cond-compiled." 2014-09-03 13:58:16 -07:00
Scott LaVarnway
dcbfacbb98 Neon version of vp8_build_intra_predictors_mby_s() and
vp8_build_intra_predictors_mbuv_s().

This patch replaces the assembly version with an intrinsic
version.

On a Nexus 7, vpxenc (in realtime mode, speed -12)
reported a performance improvement of ~2.6%.

Change-Id: I9ef65bad929450c0215253fdae1c16c8b4a8f26f
2014-09-03 13:41:27 -07:00