platform/upstream/libvpx.git
9 years agoRequire webm when explicitly requested
Johann [Fri, 30 Jan 2015 23:05:14 +0000 (15:05 -0800)]
Require webm when explicitly requested

https://code.google.com/p/webm/issues/detail?id=906

Change-Id: I72841078ff81152d21d84ccf4d5548e757685a6d

9 years agoMerge "Abort if firstpass file does not exist"
Adrian Grange [Mon, 2 Feb 2015 16:11:14 +0000 (08:11 -0800)]
Merge "Abort if firstpass file does not exist"

9 years agoMerge "Optimize coef update"
Yaowu Xu [Mon, 2 Feb 2015 04:08:29 +0000 (20:08 -0800)]
Merge "Optimize coef update"

9 years agoTry again to merge branch 'frame-parallel' into master branch.
hkuang [Tue, 27 Jan 2015 20:26:28 +0000 (12:26 -0800)]
Try again to merge branch 'frame-parallel' into master branch.

In frame parallel decode, libvpx decoder decodes several frames on all
cpus in parallel fashion. If not being flushed, it will only return frame
when all the cpus are busy. If getting flushed, it will return all the
frames in the decoder. Compare with current serial decode mode in which
libvpx decoder is idle between decode calls, libvpx decoder is busy
between decode calls.

Current frame parallel decode will only speed up the decoding for frame
parallel encoded videos. For non frame parallel encoded videos, frame
parallel decode is slower than serial decode due to lack of loopfilter
worker thread.

There are still some known issues that need to be addressed. For example:
decode frame parallel videos with segmentation enabled is not right sometimes.

* frame-parallel:
  Add error handling for frame parallel decode and unit test for that.
  Fix a bug in frame parallel decode and add a unit test for that.
  Add two test vectors to test frame parallel decode.
  Add key frame seeking to webmdec and webm_video_source.
  Implement frame parallel decode for VP9.
  Increase the thread test range to cover 5, 6, 7, 8 threads.
  Fix a bug in adding frame parallel unit test.
  Add VP9 frame-parallel unit test.
  Manually pick "Make the api behavior conform to api spec." from master branch.
  Move vp9_dec_build_inter_predictors_* to decoder folder.
  Add segmentation map array for current and last frame segmentation.
  Include the right header for VP9 worker thread.
  Move vp9_thread.* to common.
  ctrl_get_reference does not need user_priv.
  Seperate the frame buffers from VP9 encoder/decoder structure.
  Revert "Revert "Revert "Revert 3 patches from Hangyu to get Chrome to build:"""
 Conflicts:
       test/codec_factory.h
       test/decode_test_driver.cc
       test/decode_test_driver.h
       test/invalid_file_test.cc
       test/test-data.sha1
       test/test.mk
       test/test_vectors.cc
       vp8/vp8_dx_iface.c
       vp9/common/vp9_alloccommon.c
       vp9/common/vp9_entropymode.c
       vp9/common/vp9_loopfilter_thread.c
       vp9/common/vp9_loopfilter_thread.h
       vp9/common/vp9_mvref_common.c
       vp9/common/vp9_onyxc_int.h
       vp9/common/vp9_reconinter.c
       vp9/decoder/vp9_decodeframe.c
       vp9/decoder/vp9_decodeframe.h
       vp9/decoder/vp9_decodemv.c
       vp9/decoder/vp9_decoder.c
       vp9/decoder/vp9_decoder.h
       vp9/encoder/vp9_encoder.c
       vp9/encoder/vp9_pickmode.c
       vp9/encoder/vp9_rdopt.c
       vp9/vp9_cx_iface.c
       vp9/vp9_dx_iface.c

This reverts commit a18da9760a74d9ce6fb9f875706dc639c95402f5.

Change-Id: I361442ffec1586d036ea2e0ee97ce4f077585f02

9 years agovp9: rename 'near' parameters
James Zern [Fri, 30 Jan 2015 23:52:24 +0000 (15:52 -0800)]
vp9: rename 'near' parameters

+ nearest for consistency

near is a reserved word in windows builds so using it as a parameter
name may cause build failures with some configurations

Change-Id: Iddf1d4ecdb39843f14e95dbfd9dca55f07f81403

9 years agoMerge "Format fixes in vp9_rd_pick_inter_mode_sb/sub8x8"
Jingning Han [Fri, 30 Jan 2015 23:49:14 +0000 (15:49 -0800)]
Merge "Format fixes in vp9_rd_pick_inter_mode_sb/sub8x8"

9 years agoMerge "configure: echo --(disable|enable)-* cmdline options"
James Zern [Fri, 30 Jan 2015 21:16:02 +0000 (13:16 -0800)]
Merge "configure: echo --(disable|enable)-* cmdline options"

9 years agoMerge "Explicitly include vp8_rtcd.h"
Johann [Fri, 30 Jan 2015 19:19:28 +0000 (11:19 -0800)]
Merge "Explicitly include vp8_rtcd.h"

9 years agoAbort if firstpass file does not exist
Adrian Grange [Fri, 30 Jan 2015 18:42:29 +0000 (10:42 -0800)]
Abort if firstpass file does not exist

This fixes a crash if the firstpass file does not
exist when doing a two-pass encode.

Change-Id: I3a1a95d68d57125c63123d6208af7537f5a689a0

9 years agoOptimize coef update
Yaowu Xu [Thu, 29 Jan 2015 18:22:48 +0000 (10:22 -0800)]
Optimize coef update

1. move the check of search method of USE_TX_8X8 up one level to
avoid operations of build_tree_distributions()
2. count tx used and avoid computaton for coef udpate when one size
is not used at all.

Change-Id: Ia3e54a2588aa531c41377a1bfaa64385d04a592c

9 years agoEnable use_x86inc for 32bit pic enabled Darwin target
Yunqing Wang [Thu, 29 Jan 2015 19:33:50 +0000 (11:33 -0800)]
Enable use_x86inc for 32bit pic enabled Darwin target

The previous patch "Fix issues in 32bit PIC enabled build" fixed
the x86inc.asm for macho32. Now we can enable use_x86inc while
building libvpx for 32bit pic enabled Darwin target, which makes
the encoder a lot faster(>2X) in this case by turning on the
existing optimizations.

Change-Id: I5f5c7add428d73f50c935c48d0a70aed2b1eb7af

9 years agoMerge "Fix issues in 32bit PIC enabled build"
Yunqing Wang [Fri, 30 Jan 2015 00:41:25 +0000 (16:41 -0800)]
Merge "Fix issues in 32bit PIC enabled build"

9 years agoMerge "Change to update of rate control factors."
Paul Wilkins [Thu, 29 Jan 2015 21:50:52 +0000 (13:50 -0800)]
Merge "Change to update of rate control factors."

9 years agoExplicitly include vp8_rtcd.h
Johann [Thu, 29 Jan 2015 17:59:16 +0000 (09:59 -0800)]
Explicitly include vp8_rtcd.h

When referencing RTCD functions make sure the relevant
header file is included.

Change-Id: Ia0d7112d4aff9b4d8fa94648f0702371b7484031
https://code.google.com/p/webm/issues/detail?id=937

9 years agoMerge "Fix to vp9 denoiser."
Marco [Thu, 29 Jan 2015 17:10:30 +0000 (09:10 -0800)]
Merge "Fix to vp9 denoiser."

9 years agoconfigure: echo --(disable|enable)-* cmdline options
James Zern [Thu, 29 Jan 2015 04:30:51 +0000 (20:30 -0800)]
configure: echo --(disable|enable)-* cmdline options

gives a better summary of what is enabled / disabled outside of the
automatic toolchain options.

fixes issue #936

Change-Id: I1bf27593a5512713aab1473cb606c58cf3084d62

9 years agoChange to update of rate control factors.
Paul Wilkins [Tue, 27 Jan 2015 02:17:14 +0000 (18:17 -0800)]
Change to update of rate control factors.

Remove damping parameter and use the damping
formula introduced by Yaowu Xu in all cases.

Change-Id: I18db7e0d0f262d5140102f259ab07821d374d285

9 years agoSimplify update_coef_probs()
Yaowu Xu [Wed, 28 Jan 2015 23:12:42 +0000 (15:12 -0800)]
Simplify update_coef_probs()

1. reduce the size of temporaray arrays on stack
2. avoid build_tree_distribution for tx size that is not used at all.

Change-Id: I0f8d7124e16a3789d3c15ad24cf02c1c12789e2c

9 years agoFix to vp9 denoiser.
Marco [Wed, 28 Jan 2015 18:24:25 +0000 (10:24 -0800)]
Fix to vp9 denoiser.

Prevent from using wrong mv for denoiser motion compensation.

Change-Id: Ifa0f9daabdbdab0900d3c17304059fe0d15de914

9 years agoMerge "vp8enc: Prevent out of bounds memory access."
Alex Converse [Wed, 28 Jan 2015 19:22:36 +0000 (11:22 -0800)]
Merge "vp8enc: Prevent out of bounds memory access."

9 years agoMerge "Add vp9_sad32x32x4d_neon Neon intrinsic function."
Frank Galligan [Wed, 28 Jan 2015 18:35:50 +0000 (10:35 -0800)]
Merge "Add vp9_sad32x32x4d_neon Neon intrinsic function."

9 years agoMerge "Add vp9_sad16x16x4d_neon Neon intrinsic function."
Frank Galligan [Wed, 28 Jan 2015 07:01:44 +0000 (23:01 -0800)]
Merge "Add vp9_sad16x16x4d_neon Neon intrinsic function."

9 years agoMerge "Add vp9_sad64x64x4d_neon Neon intrinsic function."
Frank Galligan [Wed, 28 Jan 2015 07:01:15 +0000 (23:01 -0800)]
Merge "Add vp9_sad64x64x4d_neon Neon intrinsic function."

9 years agoFix issues in 32bit PIC enabled build
Yunqing Wang [Tue, 27 Jan 2015 05:35:07 +0000 (21:35 -0800)]
Fix issues in 32bit PIC enabled build

This patch was to fix issue 924:
https://code.google.com/p/webm/issues/detail?id=924

The SECTION_RODATA macro was modified to support macho32 format.
The sub-pixel functions were modified to pass in 2 more parameters
to handle the global offsets for PIC build.

Change-Id: I3bfcd336bcae945edf300bca4ab40376a2628cd4

9 years agovp8enc: Prevent out of bounds memory access.
Alex Converse [Sat, 17 Jan 2015 00:02:05 +0000 (16:02 -0800)]
vp8enc: Prevent out of bounds memory access.

Prevent out of bounds access when attempting to increase frame size

Change-Id: I710c40c692802a72963c9680c2125da17f9060a9

9 years agoMerge "move clear_system_state() call before using double"
Yaowu Xu [Tue, 27 Jan 2015 20:42:13 +0000 (12:42 -0800)]
Merge "move clear_system_state() call before using double"

9 years agoMerge "Fix discovery of Darwin SDKs"
Johann [Tue, 27 Jan 2015 18:49:26 +0000 (10:49 -0800)]
Merge "Fix discovery of Darwin SDKs"

9 years agoAdd vp9_sad32x32x4d_neon Neon intrinsic function.
Frank Galligan [Sun, 25 Jan 2015 00:28:20 +0000 (16:28 -0800)]
Add vp9_sad32x32x4d_neon Neon intrinsic function.

On Nexus 7 speed -6 saw ~18% increase in perf.

Tested on Nexus 7, built with ndk r10d, gcc 4.9.

BUG=https://code.google.com/p/webm/issues/detail?id=908

Change-Id: I70ccdea0326750552ed946fb004507d6efe02d5c

9 years agoAdd vp9_sad16x16x4d_neon Neon intrinsic function.
Frank Galligan [Sat, 24 Jan 2015 23:43:36 +0000 (15:43 -0800)]
Add vp9_sad16x16x4d_neon Neon intrinsic function.

On Nexus 7 speed -6 saw ~15% increase in perf.

Tested on Nexus 7, built with ndk r10d, gcc 4.9.

BUG=https://code.google.com/p/webm/issues/detail?id=908

Change-Id: I4b2006b644c488f42bf06d8a22ef0e6120a96bf9

9 years agoAdd vp9_sad64x64x4d_neon Neon intrinsic function.
Frank Galligan [Sat, 24 Jan 2015 20:11:16 +0000 (12:11 -0800)]
Add vp9_sad64x64x4d_neon Neon intrinsic function.

On Nexus 7 speed -6 saw ~30% increase in perf.

Tested on Nexus 7, built with ndk r10d, gcc 4.9.

BUG=https://code.google.com/p/webm/issues/detail?id=908

Change-Id: Id12af7d1883243c23e6692e898aea82299633d58

9 years agoMerge "aq-mode=3: Update to allow for refresh on modes other than zero-mv."
Marco [Tue, 27 Jan 2015 03:47:13 +0000 (19:47 -0800)]
Merge "aq-mode=3: Update to allow for refresh on modes other than zero-mv."

9 years agoFix discovery of Darwin SDKs
Lawrence Velázquez [Sat, 24 Jan 2015 03:30:23 +0000 (22:30 -0500)]
Fix discovery of Darwin SDKs

The current method doesn't work with Xcode 4 and up, since they no
longer have a $DEVELOPER_DIR/SDKs directory. Using xcrun and xcodebuild
works all the way back to Xcode 3 on OS X 10.6 Snow Leopard, if not
earlier.

Change-Id: I7126f2fb4a8f1d6e46f921e70bbd090f00ce3d36

9 years agomove clear_system_state() call before using double
Yaowu Xu [Mon, 26 Jan 2015 23:29:15 +0000 (15:29 -0800)]
move clear_system_state() call before using double

Floating point is used in vp9_convert_qindex_to_q(), so sometime unit
test ActiveMapTest would cause run time error without properly call
to clear_system_state to reset register status.

Change-Id: I181e9395148c44a6ca8b97d6e109bd4a152143c6

9 years agoMerge "Adjust active maxq for GF groups."
Paul Wilkins [Tue, 27 Jan 2015 02:19:09 +0000 (18:19 -0800)]
Merge "Adjust active maxq for GF groups."

9 years agoMerge "Fix MSVC warnings on conversion from int64 to int"
Yaowu Xu [Tue, 27 Jan 2015 00:52:30 +0000 (16:52 -0800)]
Merge "Fix MSVC warnings on conversion from int64 to int"

9 years agoaq-mode=3: Update to allow for refresh on modes other than zero-mv.
Marco [Thu, 22 Jan 2015 00:09:13 +0000 (16:09 -0800)]
aq-mode=3: Update to allow for refresh on modes other than zero-mv.

Add distortion threshold condition to refresh state of a coding block,
and allow for qp adjustment also for some intra modes and non-zero motion modes.

Also some code cleanup (remove unused variables/code).

Change-Id: I735fa2b28bc64f60e0323976b82510577b074203

9 years agoMerge "iosbuild.sh: Increase build speed."
Tom Finegan [Mon, 26 Jan 2015 23:10:18 +0000 (15:10 -0800)]
Merge "iosbuild.sh: Increase build speed."

9 years agoAdjust active maxq for GF groups.
Paul Wilkins [Tue, 20 Jan 2015 23:23:57 +0000 (15:23 -0800)]
Adjust active maxq for GF groups.

Currently disabled by default: enabled using
#define GROUP_ADAPTIVE_MAXQ

In this patch the active max Q is adjusted for each GF
group based on the vbr bit allocation and raw first pass
group error.

This will tend to give a lower q for easy sections
and a higher value for very hard sections. As such it is
expected to improve quality in some of the easier
sections where quality issues have been reported.

This change tends to hurt overall psnr but help
average psnr. SSIM also shows a small gain.

Average results for derf, yt, std-hd and yt-hd test sets were
as follows (%change for average psnr, overal psnr and ssim):-

derf +0.291, - 0.252, -0.021
yt +6.466, -1.436, +0.552
std-hd +0.490, +0.014, +0.380
yt-hd +5.565, - 1.573, +0.099

Change-Id: Icc015499cebbf2a45054a05e8e31f3dfb43f944a

9 years agoFix MSVC warnings on conversion from int64 to int
Yaowu Xu [Mon, 26 Jan 2015 18:54:06 +0000 (10:54 -0800)]
Fix MSVC warnings on conversion from int64 to int

Change-Id: I7e96509ffa36899fcd2935749927a1e8aac8d025

9 years agoAdd Neon intrinsic vp9_fdct8x8_quant_neon
Frank Galligan [Fri, 16 Jan 2015 03:29:46 +0000 (19:29 -0800)]
Add Neon intrinsic vp9_fdct8x8_quant_neon

On Nexus 7 speed -5 got ~2%, -6 got ~15%, -7 and -8 got ~30%
increase in perf.

Tested on Nexus 7, built with ndk r10d, gcc 4.9.

Change-Id: I83246d63b96674d170098a572fa4fe28a05aaf51

9 years agoMerge "Replace divide with look-up"
Yaowu Xu [Sat, 24 Jan 2015 05:12:18 +0000 (21:12 -0800)]
Merge "Replace divide with look-up"

9 years agox86: correct OSXSAVE + AVX bit check
James Zern [Fri, 23 Jan 2015 22:13:51 +0000 (14:13 -0800)]
x86: correct OSXSAVE + AVX bit check

the result should have both bits set; previously this was converted from
webp incorrectly and resulted in a boolean check...

Change-Id: I2a7c7f2b491945f3a536ab4fca02247eccc892b8

9 years agoFormat fixes in vp9_rd_pick_inter_mode_sb/sub8x8
Jingning Han [Fri, 23 Jan 2015 19:47:15 +0000 (11:47 -0800)]
Format fixes in vp9_rd_pick_inter_mode_sb/sub8x8

Add parentheses to bit operations.

Change-Id: I095d601f0631d055adc4b3a8fde70c9cbae9e749

9 years agoMerge "SSE2 code for the filter in MFQE."
JackyChen [Fri, 23 Jan 2015 19:08:16 +0000 (11:08 -0800)]
Merge "SSE2 code for the filter in MFQE."

9 years agoMerge "Remove elevate_newmv_thresh from SPEED_FEATURES (unused)"
Adrian Grange [Fri, 23 Jan 2015 17:57:03 +0000 (09:57 -0800)]
Merge "Remove elevate_newmv_thresh from SPEED_FEATURES (unused)"

9 years agoReplace divide with look-up
Yaowu Xu [Thu, 22 Jan 2015 23:27:43 +0000 (15:27 -0800)]
Replace divide with look-up

This commit replaces an integer divide with a table-lookup. It is
to improve decoding speed, and at the same time, to reduce possible
complications with a bug in AMD Family 12h processors:

"665 Integer Divide Instruction May Cause Unpredictable Behavior"

Change-Id: I678b707a538798a923850bac467e66e847e6def7

9 years agoMerge "Revert "Merge branch 'frame-parallel' to enable frame parallel decode in maste...
Johann [Fri, 23 Jan 2015 16:43:15 +0000 (08:43 -0800)]
Merge "Revert "Merge branch 'frame-parallel' to enable frame parallel decode in master branch.""

9 years agoRevert "Merge branch 'frame-parallel' to enable frame parallel decode in master branch."
Johann [Fri, 23 Jan 2015 16:42:02 +0000 (08:42 -0800)]
Revert "Merge branch 'frame-parallel' to enable frame parallel decode in master branch."

This reverts commit bde04ce5039cbcf86c8b34bdb4127e18d7e1d0c7

Change-Id: I053dae04c761b04a36dc239558503905a14d2470

9 years agoMerge "workaround stack bashing by asm on 32-bit OpenBSD"
James Zern [Fri, 23 Jan 2015 04:04:07 +0000 (20:04 -0800)]
Merge "workaround stack bashing by asm on 32-bit OpenBSD"

9 years agoMerge branch 'master' of ssh://gerrit.chromium.org:29418/webm/libvpx
hkuang [Fri, 23 Jan 2015 02:19:04 +0000 (18:19 -0800)]
Merge branch 'master' of ssh://gerrit.chromium.org:29418/webm/libvpx

* 'master' of ssh://gerrit.chromium.org:29418/webm/libvpx:
  Add libvpx build targets for OS X 10.10 Yosemite.

9 years agoMerge branch 'frame-parallel' to enable frame parallel decode in master branch.
hkuang [Wed, 21 Jan 2015 22:51:08 +0000 (14:51 -0800)]
Merge branch 'frame-parallel' to enable frame parallel decode in master branch.

In frame parallel decode, libvpx decoder decodes several frames on all
cpus in parallel fashion. If not being flushed, it will only return frame
when all the cpus are busy. If getting flushed, it will return all the
frames in the decoder. Compare with current serial decode mode in which
libvpx decoder is idle between decode calls, libvpx decoder is busy
between decode calls. VP9 frame parallel decode is >30% faster than serial
decode with tile parallel threading which will makes devices play 1080P
VP9 videos more easily.

* frame-parallel:
  Add error handling for frame parallel decode and unit test for that.
  Fix a bug in frame parallel decode and add a unit test for that.
  Add two test vectors to test frame parallel decode.
  Add key frame seeking to webmdec and webm_video_source.
  Implement frame parallel decode for VP9.
  Increase the thread test range to cover 5, 6, 7, 8 threads.
  Fix a bug in adding frame parallel unit test.
  Add VP9 frame-parallel unit test.
  Manually pick "Make the api behavior conform to api spec." from master branch.
  Move vp9_dec_build_inter_predictors_* to decoder folder.
  Add segmentation map array for current and last frame segmentation.
  Include the right header for VP9 worker thread.
  Move vp9_thread.* to common.
  ctrl_get_reference does not need user_priv.
  Seperate the frame buffers from VP9 encoder/decoder structure.
  Revert "Revert "Revert "Revert 3 patches from Hangyu to get Chrome to build:"""

 Conflicts:
       test/codec_factory.h
       test/decode_test_driver.cc
       test/decode_test_driver.h
       test/invalid_file_test.cc
       test/test-data.sha1
       test/test.mk
       test/test_vectors.cc
       vp8/vp8_dx_iface.c
       vp9/common/vp9_alloccommon.c
       vp9/common/vp9_entropymode.c
       vp9/common/vp9_loopfilter_thread.c
       vp9/common/vp9_loopfilter_thread.h
       vp9/common/vp9_mvref_common.c
       vp9/common/vp9_onyxc_int.h
       vp9/common/vp9_reconinter.c
       vp9/decoder/vp9_decodeframe.c
       vp9/decoder/vp9_decodeframe.h
       vp9/decoder/vp9_decodemv.c
       vp9/decoder/vp9_decoder.c
       vp9/decoder/vp9_decoder.h
       vp9/encoder/vp9_encoder.c
       vp9/encoder/vp9_pickmode.c
       vp9/encoder/vp9_rdopt.c
       vp9/vp9_cx_iface.c
       vp9/vp9_dx_iface.c

Change-Id: Ib92eb35851c172d0624970e312ed515054e5ca64

9 years agoMerge "Add libvpx build targets for OS X 10.10 Yosemite."
Johann [Fri, 23 Jan 2015 02:11:21 +0000 (18:11 -0800)]
Merge "Add libvpx build targets for OS X 10.10 Yosemite."

9 years agoRemove elevate_newmv_thresh from SPEED_FEATURES (unused)
Adrian Grange [Thu, 22 Jan 2015 22:53:18 +0000 (14:53 -0800)]
Remove elevate_newmv_thresh from SPEED_FEATURES (unused)

Change-Id: I78ef7f89586a329787f6bc4c58ec83af210989a3

9 years agoAdd libvpx build targets for OS X 10.10 Yosemite.
Lawrence Velázquez [Thu, 22 Jan 2015 21:46:02 +0000 (16:46 -0500)]
Add libvpx build targets for OS X 10.10 Yosemite.

Change-Id: I5baa4405e0b52fd3b6f312bd2dc94b19e6ff3da7

9 years agoiosbuild.sh: Increase build speed.
Tom Finegan [Thu, 22 Jan 2015 23:18:30 +0000 (15:18 -0800)]
iosbuild.sh: Increase build speed.

Disable more stuff to speed up the build, and log default configure
args in verbose mode.

Change-Id: I40e55fc5e8d2bff0262e1d6bd4a40ee2c10d2b6d

9 years agoModify variance partition selection for low resolutions.
Marco [Tue, 6 Jan 2015 01:13:13 +0000 (17:13 -0800)]
Modify variance partition selection for low resolutions.

For low spatial resolutions: bias partittion selection to smaller block sizes,
and base the variance computation on 4x4 down-sampling.

Also move the threshold computations into the choose_partitioning,
so they are computed once for each sb block.

On low-res clips (RTC_derf) PSNR/SSIMetrics increase by about 4-5%.
No change for resolutions above CIF.

Change-Id: I93f8ff742c8044786977bb6e31dcf8efda6dd1b0

9 years agoMerge "Bug when last group before forced key frame is short."
Paul Wilkins [Thu, 22 Jan 2015 16:28:19 +0000 (08:28 -0800)]
Merge "Bug when last group before forced key frame is short."

9 years agoBug when last group before forced key frame is short.
Paul Wilkins [Wed, 21 Jan 2015 19:32:27 +0000 (11:32 -0800)]
Bug when last group before forced key frame is short.

Just before a forced key frame we often get a foreshortened
arf/gf group. In such a case, we do not want to update
rc->last_boosted_qindex, which is used to define the Q range
for the forced key frame itself.

This gives a small average metrics gain for the YT and YT-HD sets
(eg. YT SSIM +0.141%).

Change-Id: Ie06698bc4f249e87183b8f8fb27ff8f3fde216d9

9 years agoMerge "Fix compile error in Chromium building."
JackyChen [Wed, 21 Jan 2015 22:52:32 +0000 (14:52 -0800)]
Merge "Fix compile error in Chromium building."

9 years agoMerge "Allow external resize via vpx_codec_enc_config_set"
Alex Converse [Wed, 21 Jan 2015 21:59:34 +0000 (13:59 -0800)]
Merge "Allow external resize via vpx_codec_enc_config_set"

9 years agoFix compile error in Chromium building.
JackyChen [Wed, 21 Jan 2015 20:59:25 +0000 (12:59 -0800)]
Fix compile error in Chromium building.

The comparison of address in the condition is not necessary, since
they will constantly be non-null.

Change-Id: Id0b0075283f5af65215d5761a8160a4cb2a15c9b

9 years agoMerge "fix AVX & AVX2 detection"
James Zern [Wed, 21 Jan 2015 20:59:07 +0000 (12:59 -0800)]
Merge "fix AVX & AVX2 detection"

9 years agoAllow external resize via vpx_codec_enc_config_set
Alex Converse [Tue, 13 Jan 2015 00:26:05 +0000 (16:26 -0800)]
Allow external resize via vpx_codec_enc_config_set

Change-Id: I3d324e2baa4de2d266c5f7ca7b635b62372e90a7

9 years agoMerge "Replace "colorspace" with "color_space""
Yaowu Xu [Wed, 21 Jan 2015 16:58:09 +0000 (08:58 -0800)]
Merge "Replace "colorspace" with "color_space""

9 years agoMerge "Add Neon intrinsics for vp9_avg_8x8_neon"
Frank Galligan [Tue, 20 Jan 2015 22:38:39 +0000 (14:38 -0800)]
Merge "Add Neon intrinsics for vp9_avg_8x8_neon"

9 years agoMerge "Add non420 code in multi-threaded loopfilter"
Yunqing Wang [Tue, 20 Jan 2015 19:45:03 +0000 (11:45 -0800)]
Merge "Add non420 code in multi-threaded loopfilter"

9 years agoMerge "Fix obvious misses in doxygen mainpage"
Yaowu Xu [Tue, 20 Jan 2015 19:27:52 +0000 (11:27 -0800)]
Merge "Fix obvious misses in doxygen mainpage"

9 years agoAdd non420 code in multi-threaded loopfilter
Yunqing Wang [Fri, 16 Jan 2015 23:47:25 +0000 (15:47 -0800)]
Add non420 code in multi-threaded loopfilter

Added non420 part back to make it consistent with single
thread code in vp9_loopfilter.c.

Change-Id: I8ca255d73bffebae294d2627d6655eafe535cb90

9 years agoMerge "vp9_ethread: add parallel loopfilter"
Yunqing Wang [Tue, 20 Jan 2015 17:27:08 +0000 (09:27 -0800)]
Merge "vp9_ethread: add parallel loopfilter"

9 years agoMerge "Add vp9_highbitdepth info in configure --help"
Yaowu Xu [Tue, 20 Jan 2015 16:55:16 +0000 (08:55 -0800)]
Merge "Add vp9_highbitdepth info in configure --help"

9 years agoSSE2 code for the filter in MFQE.
JackyChen [Tue, 13 Jan 2015 18:50:50 +0000 (10:50 -0800)]
SSE2 code for the filter in MFQE.

The SSE2 code is from VP8 MFQE, reuse it in VP9. No change on VP8
side. In our testing, we achieve 2X speed by adopting this change.

Change-Id: Ib2b14144ae57c892005c1c4b84e3379d02e56716

9 years agoFix variance Neon intrinsics > 32x32
Frank Galligan [Sat, 17 Jan 2015 03:30:45 +0000 (19:30 -0800)]
Fix variance Neon intrinsics > 32x32

The 16 bit sum vector was overflowing.

Change-Id: I0fdf38e832ee99457ec8680a92691a6175ff8c3f

9 years agovp9_ethread: add parallel loopfilter
Yunqing Wang [Tue, 6 Jan 2015 22:14:26 +0000 (14:14 -0800)]
vp9_ethread: add parallel loopfilter

1. Added row-based loopfilter in encoder;
2. Moved common multi-threaded loopfilter functions from decoder
   to common;
3. Merged multi-threaded loopfilter code, and made encoder/
   decoder call same function to reduce code duplication.

Encoder tests showed that 1% - 2% speedup was seen for good-quality
2-pass mode(at speed 3); 1% - 3% speedup using 2 threads and 4% - 6%
speedup using 4 threads were seen for real-time mode(at speed 7).

Change-Id: I8a4ac51c2ad9bab9fa7b864e90743931c53ec1c4

9 years agoMerge "Fix frame buffer swap in denoiser"
Jingning Han [Sat, 17 Jan 2015 00:58:37 +0000 (16:58 -0800)]
Merge "Fix frame buffer swap in denoiser"

9 years agoMerge "[two pass temporal svc]Fix crash issue in transcoder app caused by last fix."
Minghai Shang [Fri, 16 Jan 2015 23:00:20 +0000 (15:00 -0800)]
Merge "[two pass temporal svc]Fix crash issue in transcoder app caused by last fix."

9 years agoFix frame buffer swap in denoiser
Jingning Han [Fri, 16 Jan 2015 17:03:40 +0000 (09:03 -0800)]
Fix frame buffer swap in denoiser

This commit fixes a bug in denoiser reference frame buffer swap,
which disables frame buffer update.

Change-Id: I39a9427180fd18f9692602064ad821f7af4714c0

9 years agoAdd vp9_highbitdepth info in configure --help
Yaowu Xu [Fri, 16 Jan 2015 19:15:02 +0000 (11:15 -0800)]
Add vp9_highbitdepth info in configure --help

Change-Id: I89497a483117fc472be3a836070109498258140d

9 years agofix AVX & AVX2 detection
James Zern [Thu, 15 Jan 2015 06:51:49 +0000 (22:51 -0800)]
fix AVX & AVX2 detection

fixes issue #790 which resulted in a SIGILL on OpenBSD

code is mostly from libwebp, based on the following:

https://software.intel.com/en-us/articles/how-to-detect-new-instruction-support-in-the-4th-generation-intel-core-processor-family

Change-Id: Ida7c1a18261e98c05ed9c662068140be407ec107

9 years agoReplace "colorspace" with "color_space"
Yaowu Xu [Fri, 16 Jan 2015 01:58:47 +0000 (17:58 -0800)]
Replace "colorspace" with "color_space"

This is to make the usage of the variable name consistent across
the code base.

Change-Id: I698739e55841c59358d1c6e5cc97c96088772943

9 years ago[two pass temporal svc]Fix crash issue in transcoder app caused by last fix.
Minghai Shang [Thu, 15 Jan 2015 23:44:11 +0000 (15:44 -0800)]
[two pass temporal svc]Fix crash issue in transcoder app caused by last fix.

Change-Id: I78ecc8ec3fa3ba5f69bb23813e68a5255d0534e1

9 years agoAdd Neon intrinsics for vp9_avg_8x8_neon
Frank Galligan [Thu, 15 Jan 2015 22:36:41 +0000 (14:36 -0800)]
Add Neon intrinsics for vp9_avg_8x8_neon

On Nexus 7 speed -5, -6, -7, and -8 saw about a 1% increase
in perf for 480p. Speeds -5, -6, -7, and -8 saw about a 1.5%
increase in perf for 720p.

Tested on Nexus 7, built with ndk r10d, gcc 4.9.

Change-Id: Ibf17ebfd952a6aec941719bd8306df8ec4574bee

9 years agoFix obvious misses in doxygen mainpage
Yaowu Xu [Thu, 15 Jan 2015 22:32:42 +0000 (14:32 -0800)]
Fix obvious misses in doxygen mainpage

Change-Id: I0537769904f0853c9299b36a91391f289acf442c

9 years agoFix color_space enum comments for doxygen
Yaowu Xu [Thu, 15 Jan 2015 21:40:16 +0000 (13:40 -0800)]
Fix color_space enum comments for doxygen

Change-Id: I14f1708908a0b68b7f4c8e039dbec138c14c5ae4

9 years agoFix doxygen warning with usage.dox
Yaowu Xu [Thu, 15 Jan 2015 19:42:04 +0000 (11:42 -0800)]
Fix doxygen warning with usage.dox

Change-Id: I3ee8974a66f186fb0eb15b1078a3c7b9cbf5ec80

9 years agoRemove obsolete doxygen tags
Yaowu Xu [Thu, 15 Jan 2015 02:09:14 +0000 (18:09 -0800)]
Remove obsolete doxygen tags

Change-Id: I38a42ed1d0be4fbfce6c9f3f5d021055107933d7

9 years agoMerge "Fix comments for doxygen"
Yaowu Xu [Thu, 15 Jan 2015 18:46:20 +0000 (10:46 -0800)]
Merge "Fix comments for doxygen"

9 years agoMerge "Align thread data in vp9_ethread"
Yunqing Wang [Thu, 15 Jan 2015 17:21:31 +0000 (09:21 -0800)]
Merge "Align thread data in vp9_ethread"

9 years agoFix comments for doxygen
Yaowu Xu [Tue, 13 Jan 2015 22:32:09 +0000 (14:32 -0800)]
Fix comments for doxygen

Change-Id: Iaad3dc3cdd25275939b475706eb7d136a5a10174

9 years agoworkaround stack bashing by asm on 32-bit OpenBSD
James Zern [Thu, 15 Jan 2015 07:09:57 +0000 (23:09 -0800)]
workaround stack bashing by asm on 32-bit OpenBSD

like Darwin, disable x86inc with 32-bit PIC/shared library builds;
avoids stack smashing done by vp9_subpixel_*

workaround for issue #808

Change-Id: I3f2f16bd91822c8e8dad0242b5dcfb0e621aea97

9 years agoMerge "Clarify purpose of VP9_INVERT_TILE_DECODE_ORDER"
Yaowu Xu [Thu, 15 Jan 2015 04:25:18 +0000 (20:25 -0800)]
Merge "Clarify purpose of VP9_INVERT_TILE_DECODE_ORDER"

9 years agoMerge "Add comments for two SVC related controls"
Yaowu Xu [Thu, 15 Jan 2015 01:53:49 +0000 (17:53 -0800)]
Merge "Add comments for two SVC related controls"

9 years agoMerge "Add comments for two SVC related controls"
Yaowu Xu [Thu, 15 Jan 2015 01:53:42 +0000 (17:53 -0800)]
Merge "Add comments for two SVC related controls"

9 years agoMerge "Add comments for control VP9E_SET_TUNE_CONTENT"
Yaowu Xu [Thu, 15 Jan 2015 00:28:57 +0000 (16:28 -0800)]
Merge "Add comments for control VP9E_SET_TUNE_CONTENT"

9 years agoMerge "Add comments for VP9E_SET_FRAME_PERIODIC_BOOST"
Yaowu Xu [Thu, 15 Jan 2015 00:28:41 +0000 (16:28 -0800)]
Merge "Add comments for VP9E_SET_FRAME_PERIODIC_BOOST"

9 years agoClarify purpose of VP9_INVERT_TILE_DECODE_ORDER
Yaowu Xu [Tue, 13 Jan 2015 20:19:46 +0000 (12:19 -0800)]
Clarify purpose of VP9_INVERT_TILE_DECODE_ORDER

Change-Id: I1eb98f4576994c5c9fa5ba748d85cd186910ee4f

9 years agoAlign thread data in vp9_ethread
Yunqing Wang [Wed, 14 Jan 2015 23:51:56 +0000 (15:51 -0800)]
Align thread data in vp9_ethread

On some platforms, such as 32bit Windows and 32bit Mac, the allocated
memory isn't aligned automatically. The thread data is aligned to
ensure the correct access in SIMD code.

Change-Id: I1108c145fe982ddbd3d9324952758297120e4806

9 years agoMerge "Fix comments for encoder controls"
Yaowu Xu [Wed, 14 Jan 2015 23:32:35 +0000 (15:32 -0800)]
Merge "Fix comments for encoder controls"

9 years agoMerge "Add encoder control for setting color space"
Yaowu Xu [Wed, 14 Jan 2015 22:14:34 +0000 (14:14 -0800)]
Merge "Add encoder control for setting color space"

9 years agoMerge "Switch remaining Neon variance functions to shifts"
Frank Galligan [Wed, 14 Jan 2015 20:17:42 +0000 (12:17 -0800)]
Merge "Switch remaining Neon variance functions to shifts"

9 years agoMerge "Add 64x64 sub_pel_variance Neon function"
Frank Galligan [Wed, 14 Jan 2015 20:17:20 +0000 (12:17 -0800)]
Merge "Add 64x64 sub_pel_variance Neon function"