]> granicus.if.org Git - libvpx/log
libvpx
9 years agoMerge "iosbuild.sh: Add targets argument."
Tom Finegan [Sat, 13 Dec 2014 00:14:12 +0000 (16:14 -0800)]
Merge "iosbuild.sh: Add targets argument."

9 years agoMerge "Add support for setting byte alignment."
Frank Galligan [Fri, 12 Dec 2014 23:47:11 +0000 (15:47 -0800)]
Merge "Add support for setting byte alignment."

9 years agoMerge "Remove redundant loads on 1d16_v8 filter."
James Zern [Fri, 12 Dec 2014 22:32:52 +0000 (14:32 -0800)]
Merge "Remove redundant loads on 1d16_v8 filter."

9 years agoMerge "Remove redundant loads on 1d8_v8 filter."
James Zern [Fri, 12 Dec 2014 22:32:26 +0000 (14:32 -0800)]
Merge "Remove redundant loads on 1d8_v8 filter."

9 years agoMerge "vp9: move encoder-only member from common"
James Zern [Fri, 12 Dec 2014 22:28:55 +0000 (14:28 -0800)]
Merge "vp9: move encoder-only member from common"

9 years agoMerge "don't set INLINE to 'always_inline'"
James Zern [Fri, 12 Dec 2014 22:28:20 +0000 (14:28 -0800)]
Merge "don't set INLINE to 'always_inline'"

9 years agoMerge changes Id6421838,I37499329
James Zern [Fri, 12 Dec 2014 22:27:56 +0000 (14:27 -0800)]
Merge changes Id6421838,I37499329

* changes:
  vp9: make postproc members depend on CONFIG_VP9_POSTPROC
  vp9_postproc: remove redundant CONFIG_* checks

9 years agoMerge "Allow for 4x4 prediction blocks for key frame, speed 6."
Marco [Fri, 12 Dec 2014 22:27:31 +0000 (14:27 -0800)]
Merge "Allow for 4x4 prediction blocks for key frame, speed 6."

9 years agoMerge "vp9_loopfilter_mmx: remove some unused tables"
James Zern [Fri, 12 Dec 2014 22:25:53 +0000 (14:25 -0800)]
Merge "vp9_loopfilter_mmx: remove some unused tables"

9 years agoMerge "x86_abi_support: set LIBVPX_RAND w/vp9-postproc"
James Zern [Fri, 12 Dec 2014 22:25:30 +0000 (14:25 -0800)]
Merge "x86_abi_support: set LIBVPX_RAND w/vp9-postproc"

9 years agoiosbuild.sh: Add targets argument.
Tom Finegan [Fri, 12 Dec 2014 21:53:58 +0000 (13:53 -0800)]
iosbuild.sh: Add targets argument.

Allows override of default target list. Also added missing usage info
for --extra-configure-args, and removed last vestiges of armv6 support.

Change-Id: Ic0f14fffa0cbaea1bed371d38ff65e035bbe3273

9 years agoAdd support for setting byte alignment.
Frank Galligan [Thu, 13 Nov 2014 20:28:34 +0000 (12:28 -0800)]
Add support for setting byte alignment.

Add support for setting byte alignment on the Y, U, and V plane of the
reference buffers. The byte alignment must be a power of 2, from 32 to
1024. A value of 0 sets legacy alignment.

Change-Id: I7c1399622f7aa68e123646369216b32047dda73d

9 years agoMerge "Remove unnecessary dqcoeff memset."
James Zern [Fri, 12 Dec 2014 20:16:32 +0000 (12:16 -0800)]
Merge "Remove unnecessary dqcoeff memset."

9 years agoRemove redundant loads on 1d16_v8 filter.
Frank Galligan [Fri, 12 Dec 2014 19:48:47 +0000 (11:48 -0800)]
Remove redundant loads on 1d16_v8 filter.

This CL showed about a 3% gain in performance on some systems.

Change-Id: Id27e7e0b8e69068aa364e67859436da852669250

9 years agoRemove redundant loads on 1d8_v8 filter.
Frank Galligan [Fri, 12 Dec 2014 19:34:24 +0000 (11:34 -0800)]
Remove redundant loads on 1d8_v8 filter.

This CL showed a modest gain in performance on some systems.

Change-Id: Iad636a89a1a9804ab7a0dea302bf2c6a4d1653a4

9 years agodon't set INLINE to 'always_inline'
James Zern [Thu, 11 Dec 2014 02:42:07 +0000 (18:42 -0800)]
don't set INLINE to 'always_inline'

INLINE is used quite widely in vp9, this change improves performance
1-2% on most modern platforms.

Change-Id: I8a9974aab89fa588ea4923cc7eaf6199e344a528

9 years agovp9: move encoder-only member from common
James Zern [Thu, 11 Dec 2014 02:11:17 +0000 (18:11 -0800)]
vp9: move encoder-only member from common

allow_comp_inter_inter VP9_COMMON -> VP9_COMP

Change-Id: I6d9dc25d1cdd7e2ab62f5be69cd9fa883d21dbb6

9 years agovp9: make postproc members depend on CONFIG_VP9_POSTPROC
James Zern [Thu, 11 Dec 2014 02:12:29 +0000 (18:12 -0800)]
vp9: make postproc members depend on CONFIG_VP9_POSTPROC

Change-Id: Id64218386968cee3132269e4a0572650f20fd980

9 years agovp9_postproc: remove redundant CONFIG_* checks
James Zern [Wed, 10 Dec 2014 01:27:52 +0000 (17:27 -0800)]
vp9_postproc: remove redundant CONFIG_* checks

the entire module is wrapped in CONFIG_VP9_POSTPROC which is forcibly
enabled with CONFIG_INTERNAL_STATS

+ a similar change in vp9_alloccommon.c

Change-Id: I374993297a9fba5bef2f0b71f984eba42f0995a3

9 years agovp9_loopfilter_mmx: remove some unused tables
James Zern [Wed, 10 Dec 2014 01:57:00 +0000 (17:57 -0800)]
vp9_loopfilter_mmx: remove some unused tables

Change-Id: I964d25cc91c8e4864d73b142d9c7a1b39cb6cfbb

9 years agoMerge "vp9_dx_iface.c uses CONFIG_VP9_POSTPROC but config.h not included"
Jim Bankoski [Fri, 12 Dec 2014 19:10:17 +0000 (11:10 -0800)]
Merge "vp9_dx_iface.c uses CONFIG_VP9_POSTPROC but config.h not included"

9 years agoMerge "Adds a test to make sure encoder parms get to decoder."
Jim Bankoski [Fri, 12 Dec 2014 19:10:08 +0000 (11:10 -0800)]
Merge "Adds a test to make sure encoder parms get to decoder."

9 years agox86_abi_support: set LIBVPX_RAND w/vp9-postproc
James Zern [Wed, 10 Dec 2014 01:24:49 +0000 (17:24 -0800)]
x86_abi_support: set LIBVPX_RAND w/vp9-postproc

set LIBVPX_RAND with --enable-vp9-postproc, previously only the vp8
config was checked. this fixes the build with --disable-postproc.

Change-Id: Ia61baded6aa0e44d6443ae4a3c85915f1054f053

9 years agoMerge "Fix PICK_MODE_CONTEXT index in non-RD coding mode"
Jingning Han [Fri, 12 Dec 2014 17:16:01 +0000 (09:16 -0800)]
Merge "Fix PICK_MODE_CONTEXT index in non-RD coding mode"

9 years agovp9_dx_iface.c uses CONFIG_VP9_POSTPROC but config.h not included
Jim Bankoski [Fri, 12 Dec 2014 16:42:36 +0000 (08:42 -0800)]
vp9_dx_iface.c uses CONFIG_VP9_POSTPROC but config.h not included

Change-Id: Id316b3786214bf1028992968955da917e3f2d4a3

9 years agoFix test to call clear system state in convolve_test.
Jim Bankoski [Fri, 12 Dec 2014 14:18:56 +0000 (06:18 -0800)]
Fix test to call clear system state in convolve_test.

Assembly tests should clear system state, as we have no
expectation of proper system state in between test runs..

Change-Id: I0f591996c1f17ef2a5a8572a6b445f757223a144

9 years agoAdds a test to make sure encoder parms get to decoder.
Jim Bankoski [Fri, 12 Dec 2014 01:34:32 +0000 (17:34 -0800)]
Adds a test to make sure encoder parms get to decoder.

This is meant as a framework for testing that encode parms make it
through to the decoder.

Change-Id: Idb86ee3668b45b4e73c23c6e4daef94b0650b786

9 years agoFix PICK_MODE_CONTEXT index in non-RD coding mode
Jingning Han [Fri, 12 Dec 2014 01:17:53 +0000 (17:17 -0800)]
Fix PICK_MODE_CONTEXT index in non-RD coding mode

This commit fixes a bug in the PICK_MODE_CONTEXT index for
horizontal partition case. The compression performance change
is less than 0.01% level, since most blocks are selected to
use square block size in RTC coding mode.

Change-Id: I67effc18ae8795fccdd82a55f4efc609fa5cb3e1

9 years agoMerge "Multiframe Quality Enhancement(MFQE) in VP9."
JackyChen [Fri, 12 Dec 2014 00:24:08 +0000 (16:24 -0800)]
Merge "Multiframe Quality Enhancement(MFQE) in VP9."

9 years agoAllow for 4x4 prediction blocks for key frame, speed 6.
Marco [Thu, 11 Dec 2014 23:21:17 +0000 (15:21 -0800)]
Allow for 4x4 prediction blocks for key frame, speed 6.

For key frame under variance source partition: 4x4 prediction blocks
may be selected when variance of 8x8 block is very high (threshold is set fairly high for now).

Testing on some RTC clips shows this helps to reduce some ringing artifacts on key frame.
Encoded key frame size increases about ~10%. Key frame PSNR increases about ~0.1-0.2dB.

Change-Id: I56e203fac32ea6ef69897fb3ea269c59cb50d174

9 years agoMerge "Replace division with bit shift in choose_partitioning"
Jingning Han [Thu, 11 Dec 2014 21:30:03 +0000 (13:30 -0800)]
Merge "Replace division with bit shift in choose_partitioning"

9 years agoMerge "Re-enable 8x8 fdct/fht tests by changing tolerance"
Deb Mukherjee [Thu, 11 Dec 2014 20:29:06 +0000 (12:29 -0800)]
Merge "Re-enable 8x8 fdct/fht tests by changing tolerance"

9 years agoMerge "Corrected optimization of 8x8 DCT code"
Debargha Mukherjee [Thu, 11 Dec 2014 20:28:45 +0000 (12:28 -0800)]
Merge "Corrected optimization of 8x8 DCT code"

9 years agoRemove unnecessary dqcoeff memset.
hkuang [Thu, 11 Dec 2014 20:27:25 +0000 (12:27 -0800)]
Remove unnecessary dqcoeff memset.

dqcoeff is set to be 0 on initialization. And set back to 0 after being
used everytime.

Change-Id: I32b8e149bba40a8d707849f737a8e49a691f319c

9 years agoMerge "Refactor choose_partitioning computing scheme"
Jingning Han [Thu, 11 Dec 2014 19:14:07 +0000 (11:14 -0800)]
Merge "Refactor choose_partitioning computing scheme"

9 years agoReplace division with bit shift in choose_partitioning
Jingning Han [Thu, 11 Dec 2014 19:04:49 +0000 (11:04 -0800)]
Replace division with bit shift in choose_partitioning

This commit explicitly uses the bit shift operation instead of
division for computing block variance.

Change-Id: Id19c0ff27dd1d1ae4aceee6657e1aad0d406bd74

9 years agoCorrected optimization of 8x8 DCT code
Peter de Rivaz [Thu, 11 Dec 2014 15:54:23 +0000 (15:54 +0000)]
Corrected optimization of 8x8 DCT code

The 8x8 DCT uses a fast version whenever possible.
There was a mistake in the checking code which
meant sometimes the fast version was used when it
was not safe to do so.

Change-Id: I154c84c9e2d836764768a11082947ca30f4b5ab7
(cherry picked from commit fd05fb0c21e253b4d6f92d7e0b752850ff8ab188)

9 years agoRefactor choose_partitioning computing scheme
Jingning Han [Thu, 11 Dec 2014 17:29:36 +0000 (09:29 -0800)]
Refactor choose_partitioning computing scheme

This commit refactors the choose_partitioning function. It removes
redundant memset calls and makes the encoder to calculate
variance value per block only when it is needed. It reduces the
average runtime cost of choose_partitioning by 60%. Overall it
reduces speed -6 runtime by 2-5%.

Change-Id: I951922c50d901d0fff77a3bafc45992179bacef9

9 years agoMultiframe Quality Enhancement(MFQE) in VP9.
JackyChen [Tue, 2 Dec 2014 20:14:47 +0000 (12:14 -0800)]
Multiframe Quality Enhancement(MFQE) in VP9.

It is the first version of MFQE in VP9. There are a few TODOs included
in this version.
Usage: Add flag --enable-vp9-postproc to config the project.
In decoder, use flag --mfqe in the command line to enable
MFQE in postproc.
Note: Need to have key frame with low quality to see the effect of this
new patch. In my experiment, I fixed the qindex to 200 in key frame.

Change-Id: I021f9ce4616ed3574c81e48d968662994b56a396

9 years agoEnable neon idct tests for intrinsics
Johann [Wed, 10 Dec 2014 23:20:22 +0000 (15:20 -0800)]
Enable neon idct tests for intrinsics

Change-Id: I45d4a22f3ecb9af172e37c95f168805e492c5493

9 years agoVP9 common for ARMv8 by using NEON intrinsics 18
James Yu [Tue, 18 Feb 2014 13:56:46 +0000 (21:56 +0800)]
VP9 common for ARMv8 by using NEON intrinsics 18

Add vp9_idct32x32_add_neon.c
- vp9_idct32x32_1024_add_neon

Change-Id: Ic598b772c28bd3487a8ead7a4598a66b25f9b00f
Signed-off-by: James Yu <james.yu@linaro.org>
9 years agoVP9 common for ARMv8 by using NEON intrinsics 14
James Yu [Fri, 7 Feb 2014 17:52:15 +0000 (01:52 +0800)]
VP9 common for ARMv8 by using NEON intrinsics 14

Add vp9_idct16x16_add_neon.c
- vp9_idct16x16_256_add_neon_pass1
- vp9_idct16x16_256_add_neon_pass2
- vp9_idct16x16_10_add_neon_pass1
- vp9_idct16x16_10_add_neon_pass2

Change-Id: I54d25b54a36f4371760f54e4036693aaea40a5de
Signed-off-by: James Yu <james.yu@linaro.org>
9 years agoVP9 common for ARMv8 by using NEON intrinsics 13
James Yu [Sat, 1 Feb 2014 06:56:06 +0000 (14:56 +0800)]
VP9 common for ARMv8 by using NEON intrinsics 13

Add vp9_idct8x8_add_neon.c
- vp9_idct8x8_64_add_neon
- vp9_idct8x8_10_add_neon

Change-Id: I6ee7b4496765aa36ed52990f2ef73e9f24459610
Signed-off-by: James Yu <james.yu@linaro.org>
9 years agoVP9 common for ARMv8 by using NEON intrinsics 12
James Yu [Sat, 1 Feb 2014 06:01:05 +0000 (14:01 +0800)]
VP9 common for ARMv8 by using NEON intrinsics 12

Add vp9_idct4x4_add_neon.c
- vp9_idct4x4_16_add_neon

Change-Id: I011a96b10f1992dbd52246019ce05bae7ca8ea4f
Signed-off-by: James Yu <james.yu@linaro.org>
9 years agoVP9 common for ARMv8 by using NEON intrinsics 11
James Yu [Fri, 31 Jan 2014 05:18:15 +0000 (13:18 +0800)]
VP9 common for ARMv8 by using NEON intrinsics 11

Add vp9_idct16x16_1_add_neon.c
- vp9_idct16x16_1_add_neon

Change-Id: I7c6524024ad4cb4e66aa38f1c887e733503c39df
Signed-off-by: James Yu <james.yu@linaro.org>
9 years agoVP9 common for ARMv8 by using NEON intrinsics 10
James Yu [Thu, 30 Jan 2014 07:26:31 +0000 (15:26 +0800)]
VP9 common for ARMv8 by using NEON intrinsics 10

Add vp9_idct32x32_1_add_neon.c
- vp9_idct32x32_1_add_neon

Change-Id: If9ffe9a857228f5c67f61dc2b428b40965816eda
Signed-off-by: James Yu <james.yu@linaro.org>
9 years agoVP9 common for ARMv8 by using NEON intrinsics 09
James Yu [Thu, 30 Jan 2014 04:26:44 +0000 (12:26 +0800)]
VP9 common for ARMv8 by using NEON intrinsics 09

Add vp9_idct8x8_1_add_neon.c
- vp9_idct8x8_1_add_neon

Change-Id: I9d23e01fa96013febbf64db6c76c6c955f14e3ff
Signed-off-by: James Yu <james.yu@linaro.org>
9 years agoVP9 common for ARMv8 by using NEON intrinsics 08
James Yu [Thu, 30 Jan 2014 03:54:35 +0000 (11:54 +0800)]
VP9 common for ARMv8 by using NEON intrinsics 08

Add vp9_idct4x4_1_add_neon.c
- vp9_idct4x4_1_add_neon

Change-Id: Ieab9af107dbd07a4f9503bc945890c90faccb8ac
Signed-off-by: James Yu <james.yu@linaro.org>
9 years agoMerge "VP9 common for ARMv8 by using NEON intrinsics 07"
Johann [Wed, 10 Dec 2014 19:40:46 +0000 (11:40 -0800)]
Merge "VP9 common for ARMv8 by using NEON intrinsics 07"

9 years agoMerge "VP9 common for ARMv8 by using NEON intrinsics 04"
Johann [Wed, 10 Dec 2014 19:40:29 +0000 (11:40 -0800)]
Merge "VP9 common for ARMv8 by using NEON intrinsics 04"

9 years agoMerge "Substantial restructuring of AQ mode 2."
Paul Wilkins [Wed, 10 Dec 2014 18:44:27 +0000 (10:44 -0800)]
Merge "Substantial restructuring of AQ mode 2."

9 years agoMerge "Use use_prev_frame_mvs flag for ref mv search branch"
Jingning Han [Wed, 10 Dec 2014 17:25:12 +0000 (09:25 -0800)]
Merge "Use use_prev_frame_mvs flag for ref mv search branch"

9 years agoMerge "Refactor update_state_rt"
Jingning Han [Wed, 10 Dec 2014 17:25:05 +0000 (09:25 -0800)]
Merge "Refactor update_state_rt"

9 years agoMerge "Make RTC coding flow support sub8x8 in key frame coding"
Jingning Han [Wed, 10 Dec 2014 17:24:56 +0000 (09:24 -0800)]
Merge "Make RTC coding flow support sub8x8 in key frame coding"

9 years agoMerge "Cosmetic naming change"
Jingning Han [Wed, 10 Dec 2014 17:05:34 +0000 (09:05 -0800)]
Merge "Cosmetic naming change"

9 years agoMerge "Take out redundant setting of mode_info from set_block_size"
Jingning Han [Wed, 10 Dec 2014 17:05:26 +0000 (09:05 -0800)]
Merge "Take out redundant setting of mode_info from set_block_size"

9 years agoMerge "Remove unused rd cost calculation from nonrd_use_partition"
Jingning Han [Wed, 10 Dec 2014 17:05:18 +0000 (09:05 -0800)]
Merge "Remove unused rd cost calculation from nonrd_use_partition"

9 years agoMerge changes I92251a8b,I5d23a685
Jim Bankoski [Wed, 10 Dec 2014 14:42:08 +0000 (06:42 -0800)]
Merge changes I92251a8b,I5d23a685

* changes:
  Adds a decode perf test that builds a new file.
  Make the decoder Cfg available to encoder tests..

9 years agoVP9 common for ARMv8 by using NEON intrinsics 07
James Yu [Wed, 29 Jan 2014 15:12:41 +0000 (23:12 +0800)]
VP9 common for ARMv8 by using NEON intrinsics 07

Add vp9_convolve8_neon.c
- vp9_convolve8_horiz_neon
- vp9_convolve8_vert_neon

Change-Id: I0bdd99ff72d275223fe211ac7243c25a5a60cf87
Signed-off-by: James Yu <james.yu@linaro.org>
9 years agoVP9 common for ARMv8 by using NEON intrinsics 04
James Yu [Sat, 25 Jan 2014 12:51:49 +0000 (20:51 +0800)]
VP9 common for ARMv8 by using NEON intrinsics 04

Add vp9_convolve8_avg_neon.c
- vp9_convolve8_avg_horiz_neon
- vp9_convolve8_avg_vert_neon

Change-Id: I617971e37b02186fec5aca181f4f9622050ea2df
Signed-off-by: James Yu <james.yu@linaro.org>
9 years agoVP9 common for ARMv8 by using NEON intrinsics 03
James Yu [Tue, 21 Jan 2014 09:23:27 +0000 (17:23 +0800)]
VP9 common for ARMv8 by using NEON intrinsics 03

Add vp9_copy_neon.c
- vp9_convolve_copy_neon

Change-Id: I291fc5423d06240876411bbceab03eae5ef585be
Signed-off-by: James Yu <james.yu@linaro.org>
9 years agoVP9 common for ARMv8 by using NEON intrinsics 02
Scott LaVarnway [Wed, 10 Sep 2014 16:49:34 +0000 (09:49 -0700)]
VP9 common for ARMv8 by using NEON intrinsics 02

Add vp9_avg_neon.c
- vp9_convolve_avg_neon

Change-Id: Id2c9d5bcfa37cff1a16417aba1656ff07bdf10fd
Signed-off-by: James Yu <james.yu@linaro.org>
9 years agoMerge "Fix clang ioc warning due to NULL src_mi pointer."
James Zern [Wed, 10 Dec 2014 02:31:46 +0000 (18:31 -0800)]
Merge "Fix clang ioc warning due to NULL src_mi pointer."

9 years agoUse use_prev_frame_mvs flag for ref mv search branch
Jingning Han [Wed, 10 Dec 2014 02:01:17 +0000 (18:01 -0800)]
Use use_prev_frame_mvs flag for ref mv search branch

Replace error_resilient flag with use_prev_frame_mvs in
vp9_pick_inter_mode reference motion vector search selection.
This effectively turns off the simplified ref mv search in the
settings of frame resizing, even if error-resilient mode is off.

Change-Id: I7fed814ee7bc0cb419a03b846e0fc2de46ba7686

9 years agoMerge "Add convolve_copy and convolve_avg to the test"
Johann [Wed, 10 Dec 2014 00:51:35 +0000 (16:51 -0800)]
Merge "Add convolve_copy and convolve_avg to the test"

9 years agoRefactor update_state_rt
Jingning Han [Tue, 9 Dec 2014 20:09:36 +0000 (12:09 -0800)]
Refactor update_state_rt

Update the frame motion vector only if previous frame motion vector
is needed for next frame reference motion vector.

Change-Id: Ica50f9d7b46ad4f815bba0d9e30f5546df29546f

9 years agoFix clang ioc warning due to NULL src_mi pointer.
hkuang [Tue, 9 Dec 2014 22:32:48 +0000 (14:32 -0800)]
Fix clang ioc warning due to NULL src_mi pointer.

The warning only happens in VP9 encoder's first pass due to src_mi
is not set up yet. But it will not fail the encoder as left_mi and
above_mi are not used in the first_pass and they will be set up again
in the second pass.

Change-Id: I12dffcd5fb1002b2b2dabb083c8726650e4b5f08

9 years agoMerge "VP9 common for ARMv8 by using NEON intrinsics 01"
Johann [Tue, 9 Dec 2014 21:41:49 +0000 (13:41 -0800)]
Merge "VP9 common for ARMv8 by using NEON intrinsics 01"

9 years agoAdd convolve_copy and convolve_avg to the test
Johann [Tue, 9 Dec 2014 20:05:15 +0000 (12:05 -0800)]
Add convolve_copy and convolve_avg to the test

Change-Id: Ic9438031282e63e627550f7e4cdeda36e43e647b

9 years agoMerge "Disable neon assembly when neon is disabled"
Johann [Tue, 9 Dec 2014 20:47:12 +0000 (12:47 -0800)]
Merge "Disable neon assembly when neon is disabled"

9 years agoAdds a decode perf test that builds a new file.
Jim Bankoski [Tue, 9 Dec 2014 20:44:45 +0000 (12:44 -0800)]
Adds a decode perf test that builds a new file.

This allows us to track decode speed for new encodes so that we catch
problems like an encode change that makes decode really slow.

Change-Id: I92251a8b1f710b241f66e1042413df1b71b76038

9 years agoVP9 common for ARMv8 by using NEON intrinsics 01
James Yu [Tue, 21 Jan 2014 01:43:29 +0000 (09:43 +0800)]
VP9 common for ARMv8 by using NEON intrinsics 01

Add vp9_loopfilter_neon.c
- vp9_lpf_horizontal_4_neon
- vp9_lpf_vertical_4_neon
- vp9_lpf_horizontal_8_neon
- vp9_lpf_vertical_8_neon

Change-Id: I97a0d7b399a431c21ee77396be3d5f5a1f7ebccb
Signed-off-by: James Yu <james.yu@linaro.org>
9 years agoMake RTC coding flow support sub8x8 in key frame coding
Jingning Han [Tue, 9 Dec 2014 19:31:45 +0000 (11:31 -0800)]
Make RTC coding flow support sub8x8 in key frame coding

This commit enables the use of sub8x8 blocks in RTC key frame
encoding. It requires the block size to be preset and will decide
the coding mode and encode the bit-stream.

Change-Id: I35aaf8ee2d4d6085432410c7963f339f85a2c19b

9 years agoDisable neon assembly when neon is disabled
Johann [Mon, 8 Dec 2014 23:13:37 +0000 (15:13 -0800)]
Disable neon assembly when neon is disabled

Change-Id: Idde266cd7287bb6bee016c90efeafa67550f94c6

9 years agoCosmetic naming change
Jingning Han [Tue, 9 Dec 2014 18:30:39 +0000 (10:30 -0800)]
Cosmetic naming change

Rename set_modeinfo_offsets as set_mode_info_offsets, to be more
consistent with naming convention.

Change-Id: I68ca1f36c4a78127d9439a50c1506a2afd07927d

9 years agoTake out redundant setting of mode_info from set_block_size
Jingning Han [Tue, 9 Dec 2014 18:24:37 +0000 (10:24 -0800)]
Take out redundant setting of mode_info from set_block_size

The later encoding process will take the top-left block's
mode_info for pre-determined block size.

Change-Id: I76a90f9ce7f3b2dbc2975b52442114e461c465b5

9 years agoMerge "Clean up the logic of handling corrupted frame."
hkuang [Tue, 9 Dec 2014 18:23:18 +0000 (10:23 -0800)]
Merge "Clean up the logic of handling corrupted frame."

9 years agoSubstantial restructuring of AQ mode 2.
Paul Wilkins [Thu, 27 Nov 2014 10:50:56 +0000 (10:50 +0000)]
Substantial restructuring of AQ mode 2.

The restructure moves the decision into the rd pick
modes loop and makes a decision based at the 16x16
block level instead of only the 64x64 level.

This gives finer granularity and better visual results
on the clips I have tested. Metrics results are worse
than the old AQ2 especially for PSNR and this mode
now falls between AQ0 and AQ1 in terms of visual
impact and metrics results.

Further tuning of this to follow.

It should be noted that if there are multiple iterations
of the recode loop the segment for a MB could change
in each loop if the previous loop causes a change in the
complexity / variance bin of the block. Also where a block
gets a delta Q this will alter the rd multiplier for this block
in subsequent recode iterations and frames where the
segmentation is applied.

Change-Id: I20256c125daa14734c16f7cc9aefab656ab808f7

9 years agoRemove unused rd cost calculation from nonrd_use_partition
Jingning Han [Tue, 9 Dec 2014 02:43:36 +0000 (18:43 -0800)]
Remove unused rd cost calculation from nonrd_use_partition

The per block rd cost calculation is not needed when partition
size is preset.

Change-Id: Ie5575248bbffb584e908aa13097f697ace6ec747

9 years agoMerge "Extend x32 check by also checking for __x86_64__."
Johann [Mon, 8 Dec 2014 22:52:31 +0000 (14:52 -0800)]
Merge "Extend x32 check by also checking for __x86_64__."

9 years agoMerge "SSSE3 Optimization for Atom processors using new instruction selection and...
Yunqing Wang [Mon, 8 Dec 2014 21:34:53 +0000 (13:34 -0800)]
Merge "SSSE3 Optimization for Atom processors using new instruction selection and ordering"

9 years agoMerge "Changes to assembler for NASM on mac."
James Zern [Mon, 8 Dec 2014 20:55:06 +0000 (12:55 -0800)]
Merge "Changes to assembler for NASM on mac."

9 years agoSSSE3 Optimization for Atom processors using new instruction selection and ordering
levytamar82 [Fri, 5 Dec 2014 18:14:33 +0000 (11:14 -0700)]
SSSE3 Optimization for Atom processors using new instruction selection and ordering

The function vp9_filter_block1d16_h8_ssse3 uses the PSHUFB instruction which has a 3 cycle latency and slows execution when done in blocks of 5 or more on Atom processors.
By replacing the PSHUFB instructions with other more efficient single cycle instructions (PUNPCKLBW + PUNPCHBW + PALIGNR) performance can be improved.
In the original code, the PSHUBF uses every byte and is consecutively copied.
This is done more efficiently by PUNPCKLBW and PUNPCHBW, using PALIGNR to concatenate the intermediate result and then shift right the next consecutive 16 bytes for the final result.

For example:
filter = 0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,8
Reg = 0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15
REG1 = PUNPCKLBW Reg, Reg = 0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7
REG2 = PUNPCHBW Reg, Reg = 8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15
PALIGNR REG2, REG1, 1 = 0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,8

This optimization improved the function performance by 23% and produced a 3% user level gain on 1080p content on Atom processors.
There was no observed performance impact on Core processors (expected).

Change-Id: I3cec701158993d95ed23ff04516942b5a4a461c0

9 years agoMerge "Improve the performance by caching the left_mi and right_mi in macroblockd."
hkuang [Mon, 8 Dec 2014 18:24:17 +0000 (10:24 -0800)]
Merge "Improve the performance by caching the left_mi and right_mi in macroblockd."

9 years agoMerge "Use average mb energy from first pass in AQ2 test."
Paul Wilkins [Mon, 8 Dec 2014 17:01:39 +0000 (09:01 -0800)]
Merge "Use average mb energy from first pass in AQ2 test."

9 years agoMerge "Fix potential integer overflow."
Frank Galligan [Mon, 8 Dec 2014 05:37:39 +0000 (21:37 -0800)]
Merge "Fix potential integer overflow."

9 years agoMake the decoder Cfg available to encoder tests..
Jim Bankoski [Sun, 7 Dec 2014 19:28:51 +0000 (11:28 -0800)]
Make the decoder Cfg available to encoder tests..

Adds decoder config as a changeable parameter to unit tests, and
changes end to end test to use commonly used parameters to enable
base test of tiles encoding and frame parallel decoding.

Change-Id: I5d23a6857303b4d68b92b15c3f2f04a1bcb4c2bb

9 years agoMerge "vp9 asserts: fix compile warning"
James Zern [Sat, 6 Dec 2014 05:09:42 +0000 (21:09 -0800)]
Merge "vp9 asserts: fix compile warning"

9 years agoMerge "fix building with --disable-spatial-resampling"
James Zern [Sat, 6 Dec 2014 04:36:44 +0000 (20:36 -0800)]
Merge "fix building with --disable-spatial-resampling"

9 years agofix building with --disable-spatial-resampling
James Zern [Sat, 6 Dec 2014 00:02:42 +0000 (16:02 -0800)]
fix building with --disable-spatial-resampling

vpx_scale.c is only used by the vp8 encoder when spatial resampling is
enabled.

Change-Id: If3d3ad81e9ee6e0b59f8c040b9624ef52598fe03

9 years agoMerge "vp9: fix frame-parallel encoding"
James Zern [Sat, 6 Dec 2014 03:00:48 +0000 (19:00 -0800)]
Merge "vp9: fix frame-parallel encoding"

9 years agoMerge "Some internal-stats, vp9-highbitdepth bug fixes"
Deb Mukherjee [Sat, 6 Dec 2014 01:49:52 +0000 (17:49 -0800)]
Merge "Some internal-stats, vp9-highbitdepth bug fixes"

9 years agovp9: fix frame-parallel encoding
James Zern [Sat, 6 Dec 2014 01:44:46 +0000 (17:44 -0800)]
vp9: fix frame-parallel encoding

the flag in the header wasn't being set based on the encoder
configuration in non-intra only mode

broken since:
fbc2fbf Adding oxcf temp variable.

Change-Id: Ib4cff9901889824bc4e68d7f0f6deb1e41df2f53

9 years agoMerge "Remove redundant rdcost reset"
Jingning Han [Sat, 6 Dec 2014 01:35:07 +0000 (17:35 -0800)]
Merge "Remove redundant rdcost reset"

9 years agoMerge "Fix a motion search skip condition in vp9_pick_inter_mode"
Jingning Han [Sat, 6 Dec 2014 01:35:04 +0000 (17:35 -0800)]
Merge "Fix a motion search skip condition in vp9_pick_inter_mode"

9 years agoMerge "Remove redundant MB_MODE_INFO reset from vp9_pick_mode_inter"
Jingning Han [Sat, 6 Dec 2014 00:59:50 +0000 (16:59 -0800)]
Merge "Remove redundant MB_MODE_INFO reset from vp9_pick_mode_inter"

9 years agoImprove the performance by caching the left_mi and right_mi in macroblockd.
hkuang [Fri, 5 Dec 2014 22:39:21 +0000 (14:39 -0800)]
Improve the performance by caching the left_mi and right_mi in macroblockd.

This improve the deocde performance by ~2% on Nexus 7 2013.

Change-Id: Ie9c4ba0371a149eb7fddc687a6a291c17298d6c3

9 years agovp9 asserts: fix compile warning
James Zern [Sat, 6 Dec 2014 00:20:42 +0000 (16:20 -0800)]
vp9 asserts: fix compile warning

string literal to int within an assert

Change-Id: I76a173f96b9add5bf27c3f5ad5d72c6f30e51629

9 years agoRemove redundant rdcost reset
Jingning Han [Sat, 6 Dec 2014 00:04:22 +0000 (16:04 -0800)]
Remove redundant rdcost reset

The initial reset of this_rdc in vp9_pick_inter_mode is not needed,
since it will be re-assign when used.

Change-Id: Ic0e12d741cbab292fc214c1eabb48b129af7839b

9 years agoFix a motion search skip condition in vp9_pick_inter_mode
Jingning Han [Fri, 5 Dec 2014 23:56:39 +0000 (15:56 -0800)]
Fix a motion search skip condition in vp9_pick_inter_mode

Compare the current best mode rate-distortion cost with the skip
threshold to decide if performing motion search.

Change-Id: Ia071824f8dd3b7db485f424692a485a2da6a1a9f