]>
granicus.if.org Git - libvpx/log
Alex Converse [Wed, 27 Apr 2016 23:47:30 +0000 (23:47 +0000)]
Merge "convolve_test: Avoid #if inside INSTANTIATE_TEST_CASE_P" into nextgenv2
Alex Converse [Wed, 27 Apr 2016 23:02:39 +0000 (23:02 +0000)]
Merge "buf_ans: Misc cleanup." into nextgenv2
Hui Su [Wed, 27 Apr 2016 22:00:22 +0000 (22:00 +0000)]
Merge "ext-intra: completely remove floating point operations" into nextgenv2
Alex Converse [Wed, 27 Apr 2016 01:09:40 +0000 (18:09 -0700)]
convolve_test: Avoid #if inside INSTANTIATE_TEST_CASE_P
BUG=https://bugs.chromium.org/p/webm/issues/detail?id=1200
Change-Id: I7f7754e7d1288b103a4510303d10afc68a7d8ca8
Alex Converse [Wed, 27 Apr 2016 21:38:30 +0000 (21:38 +0000)]
Merge "Fix vp10 txfm on MSVC 2015." into nextgenv2
Alex Converse [Wed, 27 Apr 2016 21:03:04 +0000 (14:03 -0700)]
buf_ans: Misc cleanup.
Change-Id: I18a3ef2ee6cdda57abcd27683b30b4e3136182c0
Debargha Mukherjee [Wed, 27 Apr 2016 20:43:32 +0000 (20:43 +0000)]
Merge "Turn skip recode off temporarily for ref-mv" into nextgenv2
Alex Converse [Wed, 27 Apr 2016 00:08:18 +0000 (17:08 -0700)]
Fix vp10 txfm on MSVC 2015.
BUG=https://bugs.chromium.org/p/webm/issues/detail?id=1187
Change-Id: Ied6d3d003ed6ab9cf4f03cdd1d0037ae755254f4
hui su [Tue, 26 Apr 2016 16:36:11 +0000 (09:36 -0700)]
ext-intra: completely remove floating point operations
No performance changes
Change-Id: Ia489041253423ddf8ebc7e2d41fbfb9e138109f0
Jingning Han [Wed, 27 Apr 2016 17:15:16 +0000 (17:15 +0000)]
Merge "Fix compound mv costing for ref-mv." into nextgenv2
Debargha Mukherjee [Tue, 26 Apr 2016 15:24:01 +0000 (08:24 -0700)]
Turn skip recode off temporarily for ref-mv
To fix tests in VP10/AltRefForcedKeyTestLarge.Frame1IsKey/*
Change-Id: I2f4f9fea515c9935d57006b709a9dd524f174b25
Debargha Mukherjee [Wed, 27 Apr 2016 16:36:24 +0000 (16:36 +0000)]
Merge "Initialize dummy variables." into nextgenv2
Geza Lore [Wed, 27 Apr 2016 12:37:29 +0000 (13:37 +0100)]
Fix compound mv costing for ref-mv.
I believe this is necessary for computing the correct rate,
when not doing joint_motion_search.
Change-Id: I7634d6d7a5e6f0a6998edb4d577dd047d80df3c8
Geza Lore [Wed, 27 Apr 2016 12:31:52 +0000 (13:31 +0100)]
Initialize dummy variables.
Valgrind flags these up as needed by handle_inter_mode.
Initializing fixes some assertion failures in the unit tests with
only ref-mv enabled.
Change-Id: I4d56c356692745dbecd9f790cdbb8dbfbaf72d55
Geza Lore [Wed, 27 Apr 2016 10:08:41 +0000 (11:08 +0100)]
Fix ext-tile without ext-partition.
Default case (when ext-partition was not configured) was incorrect
in encoder tile size initialization.
BUG=https://bugs.chromium.org/p/webm/issues/detail?id=1197
Change-Id: Ibe57cb1dc16b9fa300573816fc16d2d2f6849fc6
Yue Chen [Wed, 27 Apr 2016 06:29:38 +0000 (06:29 +0000)]
Merge "Optimization for EXT_INTER + OBMC" into nextgenv2
Yue Chen [Tue, 26 Apr 2016 22:47:43 +0000 (15:47 -0700)]
Optimization for EXT_INTER + OBMC
Remove the restriction that the neighboring predictor cannot be
used in obmc prediction if it is an interintra or wedgeinterinter
block. The inter predictor of the interintra block, or the first
inter predictor(using LAST or GOLDEN frame) of the wedgeinterinter
block will be exploited in obmc prediction.
Coding gain: 0.248% (2.833%->3.081%) lowres
Change-Id: I4ac0368b9d2f2956f266b30c1ac97db8bafa0742
Debargha Mukherjee [Tue, 26 Apr 2016 23:32:16 +0000 (23:32 +0000)]
Merge "Reduce intra transform set" into nextgenv2
Yue Chen [Tue, 26 Apr 2016 21:40:13 +0000 (21:40 +0000)]
Merge "Remove double counting for mv costs" into nextgenv2
Jingning Han [Tue, 26 Apr 2016 21:31:09 +0000 (21:31 +0000)]
Merge "Rework motion vector precision limit" into nextgenv2
Yue Chen [Tue, 26 Apr 2016 20:01:52 +0000 (13:01 -0700)]
Remove double counting for mv costs
The bug is introduced by commit
1a0352d , in which mv costs are
counted twice in joint_motion_search() in ext_inter experiment.
Change-Id: Ibace453df999d3c2e781d73f1f0912038fee2d4e
Hui Su [Tue, 26 Apr 2016 18:53:33 +0000 (18:53 +0000)]
Merge "ext-intra: get rid of some floating operations." into nextgenv2
Jingning Han [Mon, 25 Apr 2016 21:26:53 +0000 (14:26 -0700)]
Rework motion vector precision limit
This commit enables 1/8 luma component motion vector precision
for all motion vector cases. It improves the compression performance
of lowres by 0.13% and hdres by 0.49%.
Change-Id: Iccfc85e8ee1c0154dfbd18f060344f1e3db5dc18
Debargha Mukherjee [Mon, 25 Apr 2016 17:43:49 +0000 (10:43 -0700)]
Reduce intra transform set
Reduce transform set for intra for 8x8 and smalller to 7 from 12.
Also fixes an issue with prob updates.
Enocder Speed-up about 8-10%
Coding efficiency very little change.
lowres: -2.996 (from -3.055 before)
midres: -2.482 (from -2.552 before)
Change-Id: I4ba50ff967521b33c748fe423bd92f7cf4105ebc
Hui Su [Tue, 26 Apr 2016 16:41:49 +0000 (16:41 +0000)]
Merge "Keep track of zcoeff_blk in tx size/type search" into nextgenv2
hui su [Mon, 25 Apr 2016 18:44:24 +0000 (11:44 -0700)]
ext-intra: get rid of some floating operations.
No performance changes.
Change-Id: Idd4043090fec09e57520bc970ed2e39e6f7e1a5e
Debargha Mukherjee [Mon, 25 Apr 2016 21:42:23 +0000 (21:42 +0000)]
Merge "Clear X87 register state before using double." into nextgenv2
Yi Luo [Mon, 25 Apr 2016 19:43:35 +0000 (19:43 +0000)]
Merge "HBD hybrid transform 4x4 SSE4.1 optimization" into nextgenv2
Geza Lore [Mon, 25 Apr 2016 15:37:44 +0000 (16:37 +0100)]
Clear X87 register state before using double.
MMX and X87 floating point instructions cannot be mixed freely on
the 32 bit x86 architecture.
This fixes a lot of unit tests in the 32bit build with
--enable-ext-intra.
BUG=https://bugs.chromium.org/p/webm/issues/detail?id=1196
Change-Id: I0e1c3565f4b9cb4fc2d716e94d9c40e68b36fac8
Alex Converse [Mon, 25 Apr 2016 17:11:16 +0000 (17:11 +0000)]
Merge "Raise the probability resolution for rANS tokens to 10-bits per symbol" into nextgenv2
Yi Luo [Mon, 25 Apr 2016 16:41:11 +0000 (09:41 -0700)]
HBD hybrid transform 4x4 SSE4.1 optimization
- Optimization on tx_type: DCT_DCT, DCT_ADST, ADST_DCT, ADST_ADST.
- Overall encoder speed improves ~4.5%-6%.
- Update bit-exact unit test against current C version.
Change-Id: If751c030612245b1c2470200c9570cf40d655504
Jingning Han [Mon, 25 Apr 2016 16:13:04 +0000 (16:13 +0000)]
Merge "Fix out-of-bound memory access in loop filter" into nextgenv2
Jingning Han [Mon, 25 Apr 2016 16:12:51 +0000 (16:12 +0000)]
Merge "Refactor sub-pixel motion search" into nextgenv2
James Zern [Wed, 20 Apr 2016 19:44:08 +0000 (12:44 -0700)]
vp10/rdopt: quiet unused variable warning
when CONFIG_REF_MV and CONFIG_EXT_INTER are enabled
Change-Id: I17fa2b5fe0e1878333099cc5fa2b1ee36636b4d3
Yue Chen [Sat, 23 Apr 2016 16:51:07 +0000 (16:51 +0000)]
Merge "Fix EXT_INTER unit test failure in 32-bit builds" into nextgenv2
Jingning Han [Sat, 23 Apr 2016 01:10:30 +0000 (18:10 -0700)]
Fix out-of-bound memory access in loop filter
This commit fixes an out-of-bound memory access case in the
loop filter mask setting. This issue was introduced in
10232ed Refactor loopfilter level arrays to 2D.
https://chromium-review.googlesource.com/#/c/336645/
Change-Id: I7101a4a79b9ecfdd8ec5ef13a0b314cc95f48d12
Yue Chen [Sat, 23 Apr 2016 05:37:13 +0000 (22:37 -0700)]
Fix EXT_INTER unit test failure in 32-bit builds
Align new buffers that are used in interintra and wedgeinterinter prediction.
BUG=https://bugs.chromium.org/p/webm/issues/detail?id=1196
Change-Id: I1ef49fdf13c79a22cf8a1737e3d3052da0a92dfe
Jingning Han [Fri, 22 Apr 2016 22:47:03 +0000 (15:47 -0700)]
Silence compiler above-boundary warnings
Change-Id: I6d806f92e8d38d5b0b01bc8e0fd97bd8839c84df
Alex Converse [Wed, 6 Jan 2016 20:37:27 +0000 (12:37 -0800)]
Raise the probability resolution for rANS tokens to 10-bits per symbol
Change-Id: I397b5a9371c85d1df401d261143c985623e9def6
Jingning Han [Fri, 22 Apr 2016 05:22:37 +0000 (22:22 -0700)]
Refactor sub-pixel motion search
Unify the rate cost used in the motion estimation process.
Change-Id: I8e52ca9f29eee3469553433302b62fb02a038919
Jingning Han [Fri, 22 Apr 2016 04:44:55 +0000 (21:44 -0700)]
Replace left shift with multiplications
This avoids the potential risk in left shift of negative numbers.
Change-Id: I7aecb499ee6ce7342b172adc4741de5c6c107a24
Jingning Han [Fri, 22 Apr 2016 18:33:32 +0000 (11:33 -0700)]
Fix an enc/dec mismatch issue in ext-inter experiment
This commit fixes an encoding decision process issue that could
trigger enc/dec mismatch in the ext-inter experiment.
Change-Id: I6f10d1fd2fd1aa04e51df04c39a65cf72ac66c42
Yi Luo [Wed, 20 Apr 2016 01:23:17 +0000 (18:23 -0700)]
Change hybrid transform function argument from TXFM_2D_CFG* to int
Unit test shows manually developed SSE4.1 code would performs ~30%
better if TXFM_2D_CFG configuration is set in lower level. This
change only updates function signature. There is no performance
impact.
Change-Id: I62692bd50a21ffc8a944bbd6c155c0a2020ad77b
Alex Converse [Thu, 21 Apr 2016 20:27:19 +0000 (13:27 -0700)]
Unbreak the non-var_tx build.
Change-Id: I76cc3d88122de42f035fbf6508bdf3fd7c995012
Debargha Mukherjee [Thu, 21 Apr 2016 19:56:17 +0000 (19:56 +0000)]
Merge "Fix uninitialized blk_skip for VAR TX." into nextgenv2
Alex Converse [Thu, 21 Apr 2016 18:08:04 +0000 (18:08 +0000)]
Merge "Move ZERO_TOKEN into the ANS coef tokenset." into nextgenv2
Angie Chiang [Wed, 20 Apr 2016 02:17:39 +0000 (02:17 +0000)]
Merge "relax txfm test error constraint" into nextgenv2
Alex Converse [Tue, 19 Apr 2016 23:29:14 +0000 (23:29 +0000)]
Merge "Store ANS token CDFs in the FRAME_CONTEXT rather than in a global table." into nextgenv2
Alex Converse [Wed, 13 Apr 2016 18:48:37 +0000 (11:48 -0700)]
Move ZERO_TOKEN into the ANS coef tokenset.
Change-Id: I87943e027437543ab31fa3ae1aa8b2de3a063ae5
Jingning Han [Tue, 19 Apr 2016 21:20:08 +0000 (21:20 +0000)]
Merge "Handle zero motion vector residual" into nextgenv2
Hui Su [Tue, 19 Apr 2016 21:18:50 +0000 (21:18 +0000)]
Merge "Adjust optimize_b RD parameters" into nextgenv2
Hui Su [Tue, 19 Apr 2016 21:18:37 +0000 (21:18 +0000)]
Merge "Enable optimize_b for intra blocks" into nextgenv2
Angie Chiang [Tue, 19 Apr 2016 19:52:03 +0000 (19:52 +0000)]
Merge "Change the naming of txfm#d_test" into nextgenv2
Angie Chiang [Tue, 19 Apr 2016 18:59:00 +0000 (11:59 -0700)]
Change the naming of txfm#d_test
Change-Id: I151b18b38f7a000fb6e431cd42675ac4e7e9e3ca
hui su [Tue, 19 Apr 2016 17:44:27 +0000 (10:44 -0700)]
Keep track of zcoeff_blk in tx size/type search
Prevent potential problems when per transform block
zero forcing is re-enabled (a To-Do).
Change-Id: I03b0ab2a86d88058441f2ca18994cfd2e6329898
Yue Chen [Tue, 19 Apr 2016 18:41:01 +0000 (18:41 +0000)]
Merge "Remove an unsuccessful adaption of overlap sizes in obmc experiment" into nextgenv2
hui su [Mon, 18 Apr 2016 17:10:58 +0000 (10:10 -0700)]
Adjust optimize_b RD parameters
Coding gain:
lowres 0.44%
midres 0.24%
hdres 0.32%
Change-Id: Ie558203b2b2bf5c16cd49b114df3d696c4f35049
hui su [Fri, 15 Apr 2016 17:49:08 +0000 (10:49 -0700)]
Enable optimize_b for intra blocks
Coding gain:
lowres 0.05%
midres 0.10%
hdres 0.18%
Change-Id: I508b150c02588f911a8ddddfe73c770f0819fe10
Alex Converse [Fri, 19 Feb 2016 20:51:15 +0000 (12:51 -0800)]
Store ANS token CDFs in the FRAME_CONTEXT rather than in a global table.
This will facilitate bringing the zero node into the token set while
allowing its probability to vary independently.
Change-Id: I57b44c0fce44debb8e612021e44713b229d1b3cf
Alex Converse [Tue, 19 Apr 2016 16:39:18 +0000 (16:39 +0000)]
Merge "Use an exponential growth approach for the ANS reversal buffer." into nextgenv2
Geza Lore [Tue, 19 Apr 2016 13:22:48 +0000 (14:22 +0100)]
Fix uninitialized blk_skip for VAR TX.
x->blk_skip used to be uninitialized (leftover from encoding the
previous block), if cm->tx_mode != TX_MODE_SELECT (which is used with
higher --cpu-used or --rt options). This resulted in degraded coding
performance when using cm->tx_mode != TX_MODE_SELECT.
This fixes the VP10/EndToEndTestLarge.EndtoEndPSNRTest/40 unit test.
Also fixed an edge effect where encode_block in encodemb.c used the
formal width of the block (without cropping at the right edge), to
look up blk_skip, while select_tx_block in rdopt.c used the cropped
width to set blk_skip.
Change-Id: I76d0f49ac5ab3ab54203573e0d7fcfcc1c6aa10d
Yaowu Xu [Tue, 19 Apr 2016 15:43:58 +0000 (15:43 +0000)]
Merge "Merge branch 'master' into nextgenv2" into nextgenv2
Geza Lore [Tue, 19 Apr 2016 14:41:56 +0000 (15:41 +0100)]
Revert "Fix uninitialized blk_skip for VAR TX."
This reverts commit
e7b89d88354708790211ff3949fdc705a4fa1672 .
Geza Lore [Tue, 19 Apr 2016 13:22:48 +0000 (14:22 +0100)]
Fix uninitialized blk_skip for VAR TX.
x->blk_skip used to be uninitialzied (leftover from encoding the
previous block), if cm->tx_mode != TX_MODE_SELECT (which is used with
higher --cpu-used or --rt options). This resulted in degraded coding
performance when uning cm->tx_mode != TX_MODE_SELECT.
This fixes the VP10/EndToEndTestLarge.EndtoEndPSNRTest/40 unit test.
Change-Id: If39062927446798c626fc93694b4e6a4f35fa5da
Jingning Han [Thu, 14 Apr 2016 19:37:10 +0000 (12:37 -0700)]
Handle zero motion vector residual
This commit handles the zero motion vector residuals for single
and compound reference modes, respectively. It improves the coding
performance by 0.13% with no additional encoding complexity.
Change-Id: I16075a836025bd2746da2ff4698fb9261e4b08c1
Yi Luo [Mon, 18 Apr 2016 23:07:25 +0000 (23:07 +0000)]
Merge "Disable HBD 4x4 DCT_DCT HT test" into nextgenv2
Yue Chen [Mon, 18 Apr 2016 20:49:03 +0000 (13:49 -0700)]
Remove an unsuccessful adaption of overlap sizes in obmc experiment
We removed this adaption, which intended to reduce the size of
overlapped region if the neighboring block is a non-skip one. Thus,
now the width/height of the overlapping region is fixed as a half of
the current block.
Performance improvement (lowres/midres): 0.111%/0.102%
Change-Id: Ife75dad9d4eb355c78a05178b50cc015c442884f
Yaowu Xu [Mon, 18 Apr 2016 19:22:56 +0000 (12:22 -0700)]
Merge branch 'master' into nextgenv2
Conflicts:
vp10/common/scan.c
vp9/common/vp9_pred_common.c
vp9/decoder/vp9_decoder.c
Change-Id: Id559d98ea676da15d60ed464ddb6c48d3eed1111
Jingning Han [Fri, 15 Apr 2016 23:51:10 +0000 (16:51 -0700)]
Refactor transform selection process
This commit re-arranges the transform type and size selectio
process. It removes an unnecessary rate-distortion cost computation
step. Local experiments show that this speeds up the encoding
process by 6% for both the baseline and the ext-intra experiment.
Change-Id: Iab3b86a63a1e9e55548466791ed5d29a0575c1e7
Jingning Han [Mon, 18 Apr 2016 19:45:44 +0000 (19:45 +0000)]
Merge "Refactor rd_variance_adjustment function" into nextgenv2
Angie Chiang [Mon, 18 Apr 2016 19:40:33 +0000 (19:40 +0000)]
Merge changes I67543d36,I763f2924 into nextgenv2
* changes:
Reduce shift in txfm8x8
Let txfm's constant bit be the same for each stage
Yi Luo [Mon, 18 Apr 2016 19:24:31 +0000 (12:24 -0700)]
Disable HBD 4x4 DCT_DCT HT test
- HBD HT unit tests will be modified to test against new algorithm.
Change-Id: Iba58eeb21a45612685c93c98d7c846dab25e6638
Paul Wilkins [Mon, 18 Apr 2016 18:47:13 +0000 (18:47 +0000)]
Merge "Adjustment to prediction decay."
Angie Chiang [Mon, 18 Apr 2016 18:40:28 +0000 (18:40 +0000)]
Merge "Fit adst/dct's stage range into 32-bit in bd12" into nextgenv2
Angie Chiang [Mon, 18 Apr 2016 18:11:35 +0000 (18:11 +0000)]
Merge "Remove double operation from tx_size selection" into nextgenv2
Yi Luo [Mon, 18 Apr 2016 18:04:05 +0000 (18:04 +0000)]
Merge "Improvement on hybrid transform 4x4 DCT_DCT SSE4.1 optimization" into nextgenv2
Angie Chiang [Thu, 14 Apr 2016 22:16:08 +0000 (15:16 -0700)]
Remove double operation from tx_size selection
This CL fix the bug
rdopt.c:1687: choose_tx_size_from_rd: Assertion
`mbmi->tx_type == DCT_DCT' failed
It is caused by
1) mms register access before double operation
2) different compiler behaviors
code:
int64_t a = INT64_MAX;
double b = 1. * INT64_MAX;
printf("a < b: %d\n", a < b);
result:
a < b: 0
code:
--target=x86-linux-gcc
int64_t a = INT64_MAX;
double b = 1. * INT64_MAX;
printf("a < b: %d\n", a < b);
result:
a < b: 1
I remove the double operation and test it with EXT_TX experiment.
The psnr change is around 0.05%, which is considered as noise level.
Change-Id: If8935c70c8603617fcfa8571accd30ccdda786a0
Jingning Han [Fri, 15 Apr 2016 22:56:34 +0000 (15:56 -0700)]
Refactor rd_variance_adjustment function
Compute the reconstruction variance in the prediction mode search.
Change-Id: Id9c7635a9c9f5383e61c0e427e95234211834301
Yue Chen [Sun, 17 Apr 2016 18:54:33 +0000 (18:54 +0000)]
Merge "Optimization for EXT_INTER + OBMC combination" into nextgenv2
Yue Chen [Sat, 9 Apr 2016 00:28:26 +0000 (17:28 -0700)]
Optimization for EXT_INTER + OBMC combination
In the rd loop, check the perf of obmc, whose mv is copied from regular
inter predictor, when wedge interinter is better than regular inter
(previously it will force allow_obmc = 0). The condition of the early
termination before this step is relaxed to avoid skipping too many obmc
predictions. The rates of the overhead are properly calculated for these tools.
The logic of the bitstream syntax:
(a single ref) the interintra flag is sent first, only if it is 0, we
send the obmc flag;
(compound refs) the obmc flag is sent first, only if it is 0, we send
the wedge interinter flag
Coding gain
lowres: 0.428% (2.287%->2.715%)
Change-Id: I5f3a34640b398e313cbf84235c9fe2073eb2173f
Yi Luo [Fri, 15 Apr 2016 23:56:21 +0000 (23:56 +0000)]
Merge "Fix an unaligned memory allocation in HT 4x4 speed test" into nextgenv2
Angie Chiang [Fri, 15 Apr 2016 22:44:33 +0000 (22:44 +0000)]
Merge "remove redundant header" into nextgenv2
Angie Chiang [Fri, 15 Apr 2016 22:34:05 +0000 (15:34 -0700)]
remove redundant header
Change-Id: Ib0e880c341adebb238f43a6caeb661e2094e7a93
Angie Chiang [Fri, 15 Apr 2016 02:26:42 +0000 (19:26 -0700)]
relax txfm test error constraint
The error is increases because we reduce the const bit
of txfm
Change-Id: I0235a3fdb7dc6a4c0cd1c8cebb369df2a5071b94
Yi Luo [Fri, 15 Apr 2016 21:59:31 +0000 (14:59 -0700)]
Fix an unaligned memory allocation in HT 4x4 speed test
- Allocate 16-byte aligned memory.
- Disable speed test in unit tests.
Change-Id: Ibef734f4b9d39ad50e9b2e8e0a5d74565d57b409
Yi Luo [Fri, 15 Apr 2016 19:26:27 +0000 (12:26 -0700)]
Improvement on hybrid transform 4x4 DCT_DCT SSE4.1 optimization
- Implemented Angie's new fwd txfm algorithm.
- Improve ~100% than last 64-bit version; 3 times faster than
original C code.
- Passed bit-exact unit test.
Change-Id: Ica30b9768706604a6d69fe42da778441f0f5f02e
Scott LaVarnway [Fri, 15 Apr 2016 19:06:33 +0000 (19:06 +0000)]
Merge "VP9: inline vp9_get_intra_inter_context()"
Jingning Han [Fri, 15 Apr 2016 19:02:41 +0000 (12:02 -0700)]
Remove duplicated TxfmFunc declarations
Change-Id: If3876610a1fbce0988cc21ea917596bbb467df93
Zoe Liu [Fri, 15 Apr 2016 16:41:15 +0000 (16:41 +0000)]
Merge "Fix segfault with --cpu-used >= 3 and ext-refs." into nextgenv2
Johann Koenig [Fri, 15 Apr 2016 16:00:51 +0000 (16:00 +0000)]
Merge "Enable vpx_idct32x32_1024_add_neon for neon as well, not only for neon_asm"
Scott LaVarnway [Fri, 15 Apr 2016 11:58:37 +0000 (04:58 -0700)]
VP9: inline vp9_get_intra_inter_context()
Change-Id: I71366140799b9b39474b9b459082cdb250bd1905
Geza Lore [Thu, 14 Apr 2016 13:26:03 +0000 (14:26 +0100)]
Fix segfault with --cpu-used >= 3 and ext-refs.
With ext-ref enabled, it is possible that when trying to encode the
first true ALTREF frame after a keyframe, the previous ALTREF frame
(alias for the keyframe) is the same as one of the new LAST{2,3,4}
reference frames, and hence cpi->ref_frame_flags will have the ALTREF
bit clear, as computed by get_ref_frame_flags in encoder.c.
sf->alt_ref_search_fp forces the previous ALTREF frame to
be used as the only possible reference when encoding a new ALTREF
frame, but due to cpi->ref_frame_flags, some buffers will not be
initialized (see rdopt.c:7689 yv12_mb), leading to a segfault.
get_ref_frame_flags in encoder.c has been changed to prefer to keep
the LAST frame, then the ALTREF frame, then any of the LAST{2,3,4}
frames and then the GOLDEN frame in that order of preference in case
any of them are the same. This avoids the segfault and behaves the
same for the baseline.
Change-Id: I4da1991667614009da5d3061a6316c0d5dbc6c0c
Martin Storsjo [Thu, 3 Mar 2016 09:55:08 +0000 (11:55 +0200)]
Enable vpx_idct32x32_1024_add_neon for neon as well, not only for neon_asm
This was never hooked up for the 32x32_34 case as the neon_asm version
in
3f7c12da , when the intrinsics version was added.
Change-Id: Ic7db4ce5850c637315f9fe9e2de93a4f8cf9e320
Angie Chiang [Fri, 15 Apr 2016 02:12:22 +0000 (19:12 -0700)]
Reduce shift in txfm8x8
Change-Id: I67543d365cbef3c3e113f01660ae8cb744cc556d
Angie Chiang [Thu, 14 Apr 2016 22:35:14 +0000 (15:35 -0700)]
Let txfm's constant bit be the same for each stage
Change-Id: I763f2924afca526db371231bca18b38879bdf793
Angie Chiang [Mon, 4 Apr 2016 21:58:25 +0000 (14:58 -0700)]
Fit adst/dct's stage range into 32-bit in bd12
Change-Id: Ie428c6f0655873de3e77e844a2f2e4203cf47dff
Johann [Thu, 14 Apr 2016 21:47:16 +0000 (14:47 -0700)]
Apply 'const' to data not pointer
Change-Id: Ic6b695442e319f7582a7ee8e52a47ae3e38c7298
Jingning Han [Thu, 14 Apr 2016 20:55:51 +0000 (20:55 +0000)]
Merge "Clean up motion vector precision check in the encoding process" into nextgenv2
Jingning Han [Thu, 14 Apr 2016 20:55:45 +0000 (20:55 +0000)]
Merge "Apply motion vector precision check to candidate mv" into nextgenv2
Jingning Han [Thu, 14 Apr 2016 19:01:15 +0000 (19:01 +0000)]
Merge "Enable mode conversion in sub8x8 block" into nextgenv2