platform/upstream/libvpx.git
11 years agoMerge "Further improvements on the hybrid dwt/dct expt" into experimental
Deb Mukherjee [Thu, 13 Dec 2012 19:04:56 +0000 (11:04 -0800)]
Merge "Further improvements on the hybrid dwt/dct expt" into experimental

11 years agoFurther improvements on the hybrid dwt/dct expt
Deb Mukherjee [Wed, 12 Dec 2012 01:06:35 +0000 (17:06 -0800)]
Further improvements on the hybrid dwt/dct expt

Modifies the scanning pattern and uses a floating point 16x16
dct implementation for now to handle scaling better.
Also experiments are in progress with 2/6 and 9/7 wavelets.

Results have improved to within ~0.25% of 32x32 dct for std-hd
and about 0.03% for derf. This difference can probably be bridged by
re-optimizing the entropy stats for these transforms. Currently
the stats used are common between 32x32 dct and dwt/dct.

Experiments are in progress with various scan pattern - wavelet
combinations.

Ideally the subbands should be tokenized separately, and an
experiment will be condcuted next on that.

Change-Id: Ia9cbfc2d63cb7a47e562b2cd9341caf962bcc110

11 years agoMerge "New default coefficient/band probabilities." into experimental
Ronald S. Bultje [Thu, 13 Dec 2012 17:56:50 +0000 (09:56 -0800)]
Merge "New default coefficient/band probabilities." into experimental

11 years agoNew default coefficient/band probabilities.
Ronald S. Bultje [Wed, 12 Dec 2012 18:25:58 +0000 (10:25 -0800)]
New default coefficient/band probabilities.

Gives 0.5-0.6% improvement on derf and stdhd, and 1.1% on hd. The
old tables basically derive from times that we had only 4x4 or
only 4x4 and 8x8 DCTs.

Note that some values are filled with 128, because e.g. ADST ever
only occurs as Y-with-DC, as does 32x32; 16x16 ever only occurs
as Y-with-DC or as UV (as complement of 32x32 Y); and 8x8 Y2 ever
only has 4 coefficients max. If preferred, I can add values of
other tables in their place (e.g. use 4x4 2nd order high-frequency
probabilities for 8x8 2nd order), so that they make at least some
sense if we ever implement a larger 2nd order transform for the
8x8 DCT (etc.), please let me know

Change-Id: I917db356f2aff8865f528eb873c56ef43aa5ce22

11 years agoMerge "Consistently use get_prob(), clip_prob() and newly added clip_pixel()." into...
Ronald S. Bultje [Wed, 12 Dec 2012 18:34:14 +0000 (10:34 -0800)]
Merge "Consistently use get_prob(), clip_prob() and newly added clip_pixel()." into experimental

11 years agoConsistently use get_prob(), clip_prob() and newly added clip_pixel().
Ronald S. Bultje [Mon, 10 Dec 2012 20:09:07 +0000 (12:09 -0800)]
Consistently use get_prob(), clip_prob() and newly added clip_pixel().

Add a function clip_pixel() to clip a pixel value to the [0,255] range
of allowed values, and use this where-ever appropriate (e.g. prediction,
reconstruction). Likewise, consistently use the recently added function
clip_prob(), which calculates a binary probability in the [1,255] range.
If possible, try to use get_prob() or its sister get_binary_prob() to
calculate binary probabilities, for consistency.

Since in some places, this means that binary probability calculations
are changed (we use {255,256}*count0/(total) in a range of places,
and all of these are now changed to use 256*count0+(total>>1)/total),
this changes the encoding result, so this patch warrants some extensive
testing.

Change-Id: Ibeeff8d886496839b8e0c0ace9ccc552351f7628

11 years agoMerge "clean up tokenize_b() and stuff_b()" into experimental
Yaowu Xu [Tue, 11 Dec 2012 21:51:56 +0000 (13:51 -0800)]
Merge "clean up tokenize_b() and stuff_b()" into experimental

11 years agoclean up tokenize_b() and stuff_b()
Yaowu Xu [Thu, 6 Dec 2012 20:40:57 +0000 (12:40 -0800)]
clean up tokenize_b() and stuff_b()

Change-Id: I0c1be01aae933243311ad321b6c456adaec1a0f5

11 years agoMerge "experiment with CONTEXT conversion" into experimental
Yaowu Xu [Tue, 11 Dec 2012 17:46:36 +0000 (09:46 -0800)]
Merge "experiment with CONTEXT conversion" into experimental

11 years agoMerge "A bug fix related to switchable filters" into experimental
Deb Mukherjee [Mon, 10 Dec 2012 20:28:06 +0000 (12:28 -0800)]
Merge "A bug fix related to switchable filters" into experimental

11 years agoA bug fix related to switchable filters
Deb Mukherjee [Mon, 10 Dec 2012 20:10:36 +0000 (12:10 -0800)]
A bug fix related to switchable filters

The switchable count update was mistakenly inside a macro.

Change-Id: Iec04c52ad57034b88312dbaf05eee1f47ce265b3

11 years agoFurther changes to mv reference code.
Paul Wilkins [Wed, 5 Dec 2012 16:23:38 +0000 (16:23 +0000)]
Further changes to mv reference code.

Some further changes and refactoring of mv
reference code and selection of center point for
searches. Mainly relates to not passing so many
different local copies of things around.

Some place holder comments.

Change-Id: I309f10ffe9a9cde7663e7eae19eb594371c8d055

11 years agoMerge remote-tracking branch 'origin/vp9-preview' into experimental
John Koleszar [Sat, 8 Dec 2012 01:26:31 +0000 (17:26 -0800)]
Merge remote-tracking branch 'origin/vp9-preview' into experimental

11 years agoexperiment with CONTEXT conversion
Yaowu Xu [Wed, 28 Nov 2012 23:15:51 +0000 (15:15 -0800)]
experiment with CONTEXT conversion

This commit changed the ENTROPY_CONTEXT conversion between MBs that
have different transform sizes.

In additioin, this commit also did a number of cleanup/bug fix:
1. removed duplicate function vp9_fix_contexts() and changed to use
vp8_reset_mb_token_contexts() for both encoder and decoder
2. fixed a bug in stuff_mb_16x16 where wrong context was used for
the UV.
3. changed reset all context to 0 if a MB is skipped to simplify the
logic.

Change-Id: I7bc57a5fb6dbf1f85eac1543daaeb3a61633275c

11 years agolibvpx_test: ensure rtcd init functions are called
John Koleszar [Thu, 6 Dec 2012 21:56:25 +0000 (13:56 -0800)]
libvpx_test: ensure rtcd init functions are called

In addition to allowing tests to use the RTCD-enabled functions (perhaps transitively)
without having run a full encode/decode test yet, this fixes a linking issue with
Apple's G++ whereby the Common symbols (the function pointers themselves) wouldn't
be resolved. Fixing this linking issue is the primary impetus for this patch, as none
of the tests exercise the RTCD functionality except through the main API.

Change-Id: I12aed91ca37a707e5309aa6cb9c38a649c06bc6a

11 years agoMerge "Fix implicit cast." into vp9-preview
Jim Bankoski [Sat, 8 Dec 2012 01:16:01 +0000 (17:16 -0800)]
Merge "Fix implicit cast." into vp9-preview

11 years agoMerge "Fix meaninglesss if." into vp9-preview
Jim Bankoski [Sat, 8 Dec 2012 01:15:52 +0000 (17:15 -0800)]
Merge "Fix meaninglesss if." into vp9-preview

11 years agoClean up 4x4 coefficient decoding code.
Ronald S. Bultje [Sat, 8 Dec 2012 00:26:25 +0000 (16:26 -0800)]
Clean up 4x4 coefficient decoding code.

Don't use vp9_decode_coefs_4x4() for 2nd order DC or luma blocks. The
code introduces some overhead which is unnecessary for these cases.
Also, remove variable declarations that are only used once, remove
magic offsets into the coefficient buffer (use xd->block[i].qcoeff
instead of xd->qcoeff + magic_offset), and fix a few Google Style
Guide violations.

Change-Id: I0ae653fd80ca7f1e4bccd87ecef95ddfff8f28b4

11 years agoIntroduce vp9_coeff_probs/counts/stats/accum types.
Ronald S. Bultje [Sat, 8 Dec 2012 00:09:59 +0000 (16:09 -0800)]
Introduce vp9_coeff_probs/counts/stats/accum types.

Use these, instead of the 4/5-dimensional arrays, to hold statistics,
counts, accumulations and probabilities for coefficient tokens. This
commit also re-allows ENTROPY_STATS to compile.

Change-Id: If441ffac936f52a3af91d8f2922ea8a0ceabdaa5

11 years agoFix meaninglesss if.
Frank Galligan [Fri, 7 Dec 2012 23:20:39 +0000 (15:20 -0800)]
Fix meaninglesss if.

Change-Id: I0cb06d77805246fe39d39ad3bc5df3c3f52c7050

11 years agoRemove unused symbols from vp9 asm offsets C files.
Frank Galligan [Fri, 7 Dec 2012 23:29:18 +0000 (15:29 -0800)]
Remove unused symbols from vp9 asm offsets C files.

Change-Id: I366e6d175da3012f1c8607fd7fad99fbbb616091

11 years agoFix implicit cast.
Frank Galligan [Fri, 7 Dec 2012 23:25:44 +0000 (15:25 -0800)]
Fix implicit cast.

Change-Id: I1eb7433061a6c529471026e0ebdc6467942062eb

11 years ago32x32 transform for superblocks.
Ronald S. Bultje [Fri, 7 Dec 2012 22:45:05 +0000 (14:45 -0800)]
32x32 transform for superblocks.

This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds
code all over the place to wrap that in the bitstream/encoder/decoder/RD.

Some implementation notes (these probably need careful review):
- token range is extended by 1 bit, since the value range out of this
  transform is [-16384,16383].
- the coefficients coming out of the FDCT are manually scaled back by
  1 bit, or else they won't fit in int16_t (they are 17 bits). Because
  of this, the RD error scoring does not right-shift the MSE score by
  two (unlike for 4x4/8x8/16x16).
- to compensate for this loss in precision, the quantizer is halved
  also. This is currently a little hacky.
- FDCT and IDCT is double-only right now. Needs a fixed-point impl.
- There are no default probabilities for the 32x32 transform yet; I'm
  simply using the 16x16 luma ones. A future commit will add newly
  generated probabilities for all transforms.
- No ADST version. I don't think we'll add one for this level; if an
  ADST is desired, transform-size selection can scale back to 16x16
  or lower, and use an ADST at that level.

Additional notes specific to Debargha's DWT/DCT hybrid:
- coefficient scale is different for the top/left 16x16 (DCT-over-DWT)
  block than for the rest (DWT pixel differences) of the block. Therefore,
  RD error scoring isn't easily scalable between coefficient and pixel
  domain. Thus, unfortunately, we need to compute the RD distortion in
  the pixel domain until we figure out how to scale these appropriately.

Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b

11 years agolibvpx_test: ensure rtcd init functions are called
John Koleszar [Thu, 6 Dec 2012 21:56:25 +0000 (13:56 -0800)]
libvpx_test: ensure rtcd init functions are called

In addition to allowing tests to use the RTCD-enabled functions (perhaps transitively)
without having run a full encode/decode test yet, this fixes a linking issue with
Apple's G++ whereby the Common symbols (the function pointers themselves) wouldn't
be resolved. Fixing this linking issue is the primary impetus for this patch, as none
of the tests exercise the RTCD functionality except through the main API.

Change-Id: I12aed91ca37a707e5309aa6cb9c38a649c06bc6a

11 years agoMove vp8_scale_frame to vpx namespace
Johann [Tue, 4 Dec 2012 00:23:49 +0000 (16:23 -0800)]
Move vp8_scale_frame to vpx namespace

Change-Id: I92d613e89c8f1174eca0789116120bfa20c25c28

11 years agoRemove last duck_ functions
Johann [Mon, 3 Dec 2012 23:08:17 +0000 (15:08 -0800)]
Remove last duck_ functions

Change-Id: I5fbcd2006d05bfe841f3c7af9c1aeb2cb83b3149

11 years agoUse 'vpx_scale' consistently
Johann [Mon, 3 Dec 2012 22:19:49 +0000 (14:19 -0800)]
Use 'vpx_scale' consistently

Change-Id: I178352813d2b8702d081caf405de9dbad9af2cc3

11 years agoMerge "Change to MV reference search." into experimental
Paul Wilkins [Wed, 5 Dec 2012 17:14:46 +0000 (09:14 -0800)]
Merge "Change to MV reference search." into experimental

11 years agoMerge remote-tracking branch 'origin/vp9-preview' into experimental
John Koleszar [Wed, 5 Dec 2012 17:01:57 +0000 (09:01 -0800)]
Merge remote-tracking branch 'origin/vp9-preview' into experimental

11 years agoBegin to refactor vpx_scale usage in VP9
Johann [Mon, 3 Dec 2012 20:26:51 +0000 (12:26 -0800)]
Begin to refactor vpx_scale usage in VP9

Only declare the functions in vpx_scale RTCD and include the relevant
header.

Remove unused files and functions in vpx_scale to avoid wasting time
renaming. vpx_scale/win32/scaleopt.c contains functions which have not
been called in a long time but are potentially optimized.

The 'vp8' functions have not been renamed yet. That is for after the
cleanup.

Change-Id: I2c325a101d60fa9d27e7dfcd5b52a864b4a1e09c

11 years agoRemove ARM optimizations from VP9
Johann [Sun, 2 Dec 2012 22:14:00 +0000 (14:14 -0800)]
Remove ARM optimizations from VP9

Change-Id: I9f0ae635fb9a95c4aa1529c177ccb07e2b76970b

11 years agoUpdate ARM for vpx_scale changes
Johann [Fri, 30 Nov 2012 20:25:01 +0000 (12:25 -0800)]
Update ARM for vpx_scale changes

Refactor asm_offsets for vpx_scale.

Change-Id: I2db0eeb28c8e757bd033c6614a1e5319a1a204a5

11 years agoMerge remote-tracking branch 'origin/vp9-preview' into experimental
John Koleszar [Wed, 5 Dec 2012 16:41:35 +0000 (08:41 -0800)]
Merge remote-tracking branch 'origin/vp9-preview' into experimental

11 years agovp9_bilinear_filters_mmx: add missing extern specifiers
John Koleszar [Wed, 5 Dec 2012 16:27:48 +0000 (08:27 -0800)]
vp9_bilinear_filters_mmx: add missing extern specifiers

Change-Id: Ibabf18947f90cb4f45052763ebf44cfb8209bd8b

11 years agoChange to MV reference search.
Paul Wilkins [Tue, 4 Dec 2012 17:21:05 +0000 (17:21 +0000)]
Change to MV reference search.

This patch reduces the cpu cost of the MV ref
search by only allowing insert for candidates
that would be in the current top 4.

This could alter the outcome and slightly favors
near candidates which are tested first but also
limits the worst case loop count to 4 and means in
many cases it will drop out and not happen.

Change-Id: Idd795a825f9fd681f30f4fcd550c34c38939e113

11 years agoMerge "Begin to refactor vpx_scale usage in VP9" into experimental
Johann [Tue, 4 Dec 2012 23:23:42 +0000 (15:23 -0800)]
Merge "Begin to refactor vpx_scale usage in VP9" into experimental

11 years agoEnable unit tests to build with MSVC
Yaowu Xu [Tue, 4 Dec 2012 20:28:10 +0000 (12:28 -0800)]
Enable unit tests to build with MSVC

Change-Id: Ie80597221bee494679ae2ba50cca22dea2609cba

11 years agoFix the build with MSVC
Yaowu Xu [Tue, 4 Dec 2012 16:35:37 +0000 (08:35 -0800)]
Fix the build with MSVC

1. remove the dependency on non existing "vp9_temporal_filter_x86.h"
2. prefix filenames with vp9_ in obj_int_extract.bat to reflect the
change of the actual filenames.

Change-Id: Ib1b4d96ac41788f76917764a6722d8461c857302

11 years agoMerge "vp9: Fix assert check." into vp9-preview
Frank Galligan [Tue, 4 Dec 2012 01:29:46 +0000 (17:29 -0800)]
Merge "vp9: Fix assert check." into vp9-preview

11 years agomerged optimiz_b_16x16() into optmize_b()
Yaowu Xu [Mon, 3 Dec 2012 22:53:45 +0000 (14:53 -0800)]
merged optimiz_b_16x16() into optmize_b()

The commit changed the trellis quantization function optimize_b() to
work for MBs using all transform sizes, and eliminated the function
for MB using 16x16 transform only, optimize_b_16x16.

Change-Id: I3fa650587ab5198ed16315b38754783a72b33ba2

11 years agoMerge "Remove ARM optimizations from VP9" into experimental
Johann [Mon, 3 Dec 2012 21:54:38 +0000 (13:54 -0800)]
Merge "Remove ARM optimizations from VP9" into experimental

11 years agoMerge "Update ARM for vpx_scale changes" into experimental
Johann [Mon, 3 Dec 2012 21:54:04 +0000 (13:54 -0800)]
Merge "Update ARM for vpx_scale changes" into experimental

11 years agoBegin to refactor vpx_scale usage in VP9
Johann [Mon, 3 Dec 2012 20:26:51 +0000 (12:26 -0800)]
Begin to refactor vpx_scale usage in VP9

Only declare the functions in vpx_scale RTCD and include the relevant
header.

Remove unused files and functions in vpx_scale to avoid wasting time
renaming. vpx_scale/win32/scaleopt.c contains functions which have not
been called in a long time but are potentially optimized.

The 'vp8' functions have not been renamed yet. That is for after the
cleanup.

Change-Id: I2c325a101d60fa9d27e7dfcd5b52a864b4a1e09c

11 years agoRemove ARM optimizations from VP9
Johann [Sun, 2 Dec 2012 22:14:00 +0000 (14:14 -0800)]
Remove ARM optimizations from VP9

Change-Id: I9f0ae635fb9a95c4aa1529c177ccb07e2b76970b

11 years agoUpdate ARM for vpx_scale changes
Johann [Fri, 30 Nov 2012 20:25:01 +0000 (12:25 -0800)]
Update ARM for vpx_scale changes

Refactor asm_offsets for vpx_scale.

Change-Id: I2db0eeb28c8e757bd033c6614a1e5319a1a204a5

11 years agoMerge "fixes --disable-vp9-encoder" into vp9-preview
Jim Bankoski [Mon, 3 Dec 2012 20:41:30 +0000 (12:41 -0800)]
Merge "fixes --disable-vp9-encoder" into vp9-preview

11 years agofixes --disable-vp9-encoder
Jim Bankoski [Mon, 3 Dec 2012 20:21:16 +0000 (12:21 -0800)]
fixes --disable-vp9-encoder

Change-Id: I467bf0fdf3b35326bcce58d5459e6d2dbfd6c5e5

11 years agovp9: Fix assert check.
Frank Galligan [Mon, 3 Dec 2012 20:18:59 +0000 (12:18 -0800)]
vp9: Fix assert check.

Change-Id: If0cc1ab60dff6abd67dae7c7b3dc83a1afd7fe65

11 years agovp9: Remove superfluous command.
Frank Galligan [Mon, 3 Dec 2012 18:20:11 +0000 (10:20 -0800)]
vp9: Remove superfluous command.

- vpx_calloc is called on arf_not_zz above.
- Note The removed vpx_memset call had an issue with sizeof.

Change-Id: I86fd7a167d0a042e581e613e2a6c0b5e63073fc6

11 years agoMerge "Rename vpx_scale/rtcd.c to vpx_scale/vpx_scale_rtcd.c" into experimental
John Koleszar [Mon, 3 Dec 2012 17:18:45 +0000 (09:18 -0800)]
Merge "Rename vpx_scale/rtcd.c to vpx_scale/vpx_scale_rtcd.c" into experimental

11 years agoSupports inter-intra prediction with superblocks
Deb Mukherjee [Fri, 30 Nov 2012 19:46:20 +0000 (11:46 -0800)]
Supports inter-intra prediction with superblocks

Adds support for compound inter-intra prediction with superblocks.
Also, fixes a bug that disabled intra modes for superblocks.

Change-Id: I4d711317e1bc19df8c2f32dc645429f7fff31036

11 years agoRename vpx_scale/rtcd.c to vpx_scale/vpx_scale_rtcd.c
Frank Galligan [Fri, 30 Nov 2012 23:48:52 +0000 (15:48 -0800)]
Rename vpx_scale/rtcd.c to vpx_scale/vpx_scale_rtcd.c

Change-Id: I4765ca3dab880a3673c15fb56b2d912ec2ea194e

11 years agoAdds switchable filters with superblocks
Deb Mukherjee [Fri, 30 Nov 2012 00:39:15 +0000 (16:39 -0800)]
Adds switchable filters with superblocks

Allows switchbale filters to be used without mismatch when the
superblock experiment is on.

Also removes a spurious clamping code in decodemv.c which causes
rare encode/decode mismatches.

Change-Id: I809d9ee0b2859552b613500b539a615515b863ae

11 years agowarnings in various experiments
Jim Bankoski [Fri, 30 Nov 2012 15:29:43 +0000 (07:29 -0800)]
warnings in various experiments

Change-Id: Ib5106d4772450f8026f823dd743f162ab833b1d6

11 years agogoogle style guide include guards
Jim Bankoski [Fri, 30 Nov 2012 00:36:10 +0000 (16:36 -0800)]
google style guide include guards

Change-Id: I2c252f3ddcc99e96c1f5d3dab8bcb25a2a3637ea

11 years agoMerge "Further improve macroblock loop filters" into experimental
Yunqing Wang [Fri, 30 Nov 2012 00:07:14 +0000 (16:07 -0800)]
Merge "Further improve macroblock loop filters" into experimental

11 years agoMerge "Minor refactoring of superblock decoding" into experimental
Deb Mukherjee [Thu, 29 Nov 2012 23:33:42 +0000 (15:33 -0800)]
Merge "Minor refactoring of superblock decoding" into experimental

11 years agoMinor refactoring of superblock decoding
Deb Mukherjee [Thu, 29 Nov 2012 23:26:56 +0000 (15:26 -0800)]
Minor refactoring of superblock decoding

Refactoring for improved readability - no bitstream or
performance change.

Change-Id: I4488ed4715f8dbe38c66431106478669041b8b33

11 years agoa vp8 warning to boot.
Jim Bankoski [Thu, 29 Nov 2012 22:24:54 +0000 (14:24 -0800)]
a vp8 warning to boot.

Change-Id: I32ed051c8b4d3e5afa4e13bb4e3b2ca5e922ff64

11 years agointrinsic warnings begone
Jim Bankoski [Thu, 29 Nov 2012 22:14:26 +0000 (14:14 -0800)]
intrinsic warnings begone

Change-Id: I6a224c590b6a2c5b91f9084ffb8083d18223a206

11 years agolast remaining warning
Jim Bankoski [Thu, 29 Nov 2012 22:07:21 +0000 (14:07 -0800)]
last remaining warning

Change-Id: I1f49d96cdb5e342041c9a72ef31df361a1b609eb

11 years agofix implicit warnings idct etc
Jim Bankoski [Thu, 29 Nov 2012 19:23:02 +0000 (11:23 -0800)]
fix implicit warnings idct etc

Change-Id: I54a122cc8c0b6ed2dbc3c6ecfcd44736cd40b687

11 years agoAdditional warning message cleanup.
Jim Bankoski [Thu, 29 Nov 2012 18:10:51 +0000 (10:10 -0800)]
Additional warning message cleanup.

Change-Id: I429a97ac57db3de0bf67ce3f3fe0c6b409f77a9e

11 years agoMerge changes Iaa67bcf1,Ibea3bc80 into experimental
Yaowu Xu [Thu, 29 Nov 2012 17:34:10 +0000 (09:34 -0800)]
Merge changes Iaa67bcf1,Ibea3bc80 into experimental

* changes:
  more warning cleanup
  unused variables & warnings

11 years agoMerge "more unused variables." into experimental
Yaowu Xu [Thu, 29 Nov 2012 17:33:59 +0000 (09:33 -0800)]
Merge "more unused variables." into experimental

11 years agoMerge "unused variable" into experimental
Yaowu Xu [Thu, 29 Nov 2012 17:33:52 +0000 (09:33 -0800)]
Merge "unused variable" into experimental

11 years agoMerge "unused var removed" into experimental
Yaowu Xu [Thu, 29 Nov 2012 17:33:41 +0000 (09:33 -0800)]
Merge "unused var removed" into experimental

11 years agoMerge "minor fix to eob check for setting CONTEXT" into experimental
Yaowu Xu [Thu, 29 Nov 2012 17:27:00 +0000 (09:27 -0800)]
Merge "minor fix to eob check for setting CONTEXT" into experimental

11 years agominor fix to eob check for setting CONTEXT
Yaowu Xu [Thu, 29 Nov 2012 01:34:02 +0000 (17:34 -0800)]
minor fix to eob check for setting CONTEXT

Previously, the "!=" check is logically incorrect when eob is at 0 and
effective coefficient starting position is 1. This commit should have
no effect on bitstream.

Change-Id: I6ce3a847c7e72bfbe4f7c74f88e3310c6b9b6d30

11 years agomore warning cleanup
Jim Bankoski [Thu, 29 Nov 2012 17:07:12 +0000 (09:07 -0800)]
more warning cleanup

Change-Id: Iaa67bcf1e866dfe255c4e458d4e51e9c708ffcf4

11 years agounused variables & warnings
Jim Bankoski [Thu, 29 Nov 2012 17:02:47 +0000 (09:02 -0800)]
unused variables & warnings

Change-Id: Ibea3bc80eb26a975faaa60268bbc93237f82bc57

11 years agomore unused variables.
Jim Bankoski [Thu, 29 Nov 2012 16:54:59 +0000 (08:54 -0800)]
more unused variables.

Change-Id: Ibe11e9275949b26a77fa9c8ac2e7c356ae533d5d

11 years agounused variable
Jim Bankoski [Thu, 29 Nov 2012 16:51:19 +0000 (08:51 -0800)]
unused variable

Change-Id: I1302a6eaa840d419e8bb9ad0673e42ef139d3fee

11 years agounused var removed
Jim Bankoski [Thu, 29 Nov 2012 16:50:20 +0000 (08:50 -0800)]
unused var removed

Change-Id: I9d0efdff0c79ea4bdd660098106b64776bdd4483

11 years agosigned mismatch mvrefcount
Jim Bankoski [Thu, 29 Nov 2012 16:13:18 +0000 (08:13 -0800)]
signed mismatch mvrefcount

Change-Id: Ie34820c1b6eaba9cf9316415a46f48af79c41646

11 years agowarning error missing void
Jim Bankoski [Thu, 29 Nov 2012 15:47:50 +0000 (07:47 -0800)]
warning error missing void

Change-Id: I914bcc669297d3414261486bf1bfb716c2ecc804

11 years agoihtllm moves to rtcd
Jim Bankoski [Thu, 29 Nov 2012 15:19:38 +0000 (07:19 -0800)]
ihtllm moves to rtcd

clears up some warnings

Change-Id: I9899637497c6ad7519f098e055ab98580ae6d688

11 years agofix vp9_vp8 files renamed
Jim Bankoski [Thu, 29 Nov 2012 14:53:08 +0000 (06:53 -0800)]
fix vp9_vp8 files renamed

Change-Id: I20c426e91ee49666db42e20eb074095ab6b8ec5d

11 years agomore rtcd cleanup
Jim Bankoski [Thu, 29 Nov 2012 00:47:30 +0000 (16:47 -0800)]
more rtcd cleanup

Change-Id: Ieefd76e164ca4aa87597da0412977614ddfbacb7

11 years agoMerge "Fixing 8x8/4x4 ADST for intra modes with tx select" into experimental
Deb Mukherjee [Thu, 29 Nov 2012 00:59:17 +0000 (16:59 -0800)]
Merge "Fixing 8x8/4x4 ADST for intra modes with tx select" into experimental

11 years agoFixing 8x8/4x4 ADST for intra modes with tx select
Deb Mukherjee [Thu, 15 Nov 2012 23:14:38 +0000 (15:14 -0800)]
Fixing 8x8/4x4 ADST for intra modes with tx select

This patch allows use of 8x8 and 4x4 ADST correctly for Intra
16x16 modes and Intra 8x8 modes when the block size selected
is smaller than the prediction mode. Also includes some cleanups
and refactoring.

Rebase.

Change-Id: Ie3257bdf07bdb9c6e9476915e3a80183c8fa005a

11 years agoMerge "remove the vp9_default_mode_contexts_a" into experimental
Yaowu Xu [Wed, 28 Nov 2012 21:56:42 +0000 (13:56 -0800)]
Merge "remove the vp9_default_mode_contexts_a" into experimental

11 years agoremove the vp9_default_mode_contexts_a
Yaowu Xu [Wed, 28 Nov 2012 19:07:50 +0000 (11:07 -0800)]
remove the vp9_default_mode_contexts_a

Given the way mode_context is updated, the benefit of an additional
default is not signficant.

Change-Id: I67489453e8781340b18e26a1cc2f04e9221004a2

11 years agofixed includes to be fully specified
Jim Bankoski [Wed, 28 Nov 2012 18:41:40 +0000 (10:41 -0800)]
fixed includes to be fully specified

Change-Id: Ia1cce221f8511561b9cbd8edb7726fbc286ff243

11 years agoMerge "remove postproc invokes" into experimental
Jim Bankoski [Wed, 28 Nov 2012 18:30:42 +0000 (10:30 -0800)]
Merge "remove postproc invokes" into experimental

11 years agoMerge "Clamp decoded feature data" into experimental
John Koleszar [Wed, 28 Nov 2012 18:08:37 +0000 (10:08 -0800)]
Merge "Clamp decoded feature data" into experimental

11 years agoMerge "Revert "make: flatten object file directories"" into experimental
John Koleszar [Wed, 28 Nov 2012 18:08:22 +0000 (10:08 -0800)]
Merge "Revert "make: flatten object file directories"" into experimental

11 years agoremove postproc invokes
Jim Bankoski [Wed, 28 Nov 2012 18:00:25 +0000 (10:00 -0800)]
remove postproc invokes

and some miscellaneous invoke left overs

Change-Id: I63191b1bfd3bea4ce30cceaeb686ec850570fc43

11 years agoLocalize Y2 entropy coding context
Yaowu Xu [Wed, 28 Nov 2012 00:02:35 +0000 (16:02 -0800)]
Localize Y2 entropy coding context

This commit makes sure Y2 entropy coding context is always updated on
every macroblock even there is no Y2 block.

Change-Id: Ie307cfc46526efe55613be39f9f178d2531b56ba

11 years agoFurther improve macroblock loop filters
Yunqing Wang [Wed, 28 Nov 2012 03:16:32 +0000 (19:16 -0800)]
Further improve macroblock loop filters

This change included:
1. Aligned reads in vp9_mbloop_filter_vertical_edge function.
Since we actually read 16 bytes, we can align the reads to read
starting at (s - 8) instead of (s - 5).
2. Combined u, v loop filters.
3. Added 8x16 transpose.

This gave 2% decoder performance gain (tulip clip).

Change-Id: Ib14c2f1645c4a3436df17fe2f24789506bf0bb58

11 years agoremoved redundant mode_context data structures
Yaowu Xu [Tue, 27 Nov 2012 20:41:59 +0000 (12:41 -0800)]
removed redundant mode_context data structures

This commit removed a couple of redundant data structures in frame
coding contextsm, mode_context and mode_context_a, and changed to
use vp9_mode_contexts only. The switch of the context for different
frame type now relies on the switch of frame coding context between
lfc and lfc_a. This commit also removed a number of memcpy among
these redundant data structure.

Change-Id: I42e8174bd60f466b0860afc44c1263896471b0f3

11 years agoClamp decoded feature data
John Koleszar [Tue, 27 Nov 2012 19:16:15 +0000 (11:16 -0800)]
Clamp decoded feature data

Not all segment feature data elements are full-range powers of two, so
there are values that can be encoded that are invalid. Add a new function
to clamp values to the maximum allowed.

Change-Id: Ie47cb80ef2d54292e6b8db9f699c57214a915bc4

11 years agoRevert "make: flatten object file directories"
John Koleszar [Tue, 27 Nov 2012 23:54:54 +0000 (15:54 -0800)]
Revert "make: flatten object file directories"

This reverts commit b72373de79800e801ce730d24cb115daf78ae660.

Change-Id: Ic1601160e11df1a018ef12da25967cfb5eebd5ba

11 years agoAdd vp9_ prefix to all vp9 files
John Koleszar [Tue, 27 Nov 2012 21:59:17 +0000 (13:59 -0800)]
Add vp9_ prefix to all vp9 files

Support for gyp which doesn't support multiple objects in the same
static library having the same basename.

Change-Id: Ib947eefbaf68f8b177a796d23f875ccdfa6bc9dc

11 years agoMerge "Improve sad3x16 SSE2 function" into experimental
Yunqing Wang [Mon, 26 Nov 2012 18:15:35 +0000 (10:15 -0800)]
Merge "Improve sad3x16 SSE2 function" into experimental

11 years agoMerge "Modified mv prediction." into experimental
Paul Wilkins [Mon, 26 Nov 2012 17:59:24 +0000 (09:59 -0800)]
Merge "Modified  mv prediction." into experimental

11 years agoModified mv prediction.
Paul Wilkins [Mon, 26 Nov 2012 11:29:08 +0000 (11:29 +0000)]
Modified  mv prediction.

Modified the  mv_pred() fuunction that chooses a centre
point from which to start step searches to use the top
candidate vectors chosen previously.

Some gains (mainly on HD and tested with SB off).
Std_hd 0.874%, YT-hd 0.174%, YT 0.05%, Derf 0.036%

Change-Id: Ie232284f561838b8ecee0e28dcbb07a9cd46cf56

11 years agoImprove sad3x16 SSE2 function
Yunqing Wang [Wed, 21 Nov 2012 00:28:08 +0000 (16:28 -0800)]
Improve sad3x16 SSE2 function

Vp9_sad3x16_sse2() is heavily called in decoder, in which the
unaligned reads consume lots of cpu cycles. When CONFIG_SUBPELREFMV
is off, the unaligned offset is 1. In this situation,
we can adjust the src_ptr to be 4-byte aligned, and then do the
aligned reads. This reduced the reading time significantly. Tests
on 1080p clip showed over 2% decoder performance gain with
CONFIG_SUBPELREFM off.

Change-Id: I953afe3ac5406107933ef49d0b695eafba9a6507

11 years agoremove the dependency on idct.h
Yaowu Xu [Mon, 26 Nov 2012 17:10:35 +0000 (09:10 -0800)]
remove the dependency on idct.h

Change-Id: Idcf827d8ae6429ee5b673c3398f838dbeacb4e74

11 years agoMerge "removed the idct rtcd idct calls" into experimental
Jim Bankoski [Sun, 25 Nov 2012 05:38:36 +0000 (21:38 -0800)]
Merge "removed the idct rtcd idct calls" into experimental