platform/upstream/libvpx.git
13 years agoMerge remote branch 'internal/upstream' into HEAD
John Koleszar [Tue, 31 Aug 2010 19:56:52 +0000 (15:56 -0400)]
Merge remote branch 'internal/upstream' into HEAD

Conflicts:
configure

13 years agofollowup arm patch
Johann [Thu, 26 Aug 2010 20:11:30 +0000 (16:11 -0400)]
followup arm patch

make the arm asm detokenizer work with the new structures

Change-Id: I7cd92c2a018ec24032bb1cfd1bb9739bc84b444a

13 years agoChanged above and left context data layout
Scott LaVarnway [Tue, 31 Aug 2010 14:49:57 +0000 (10:49 -0400)]
Changed above and left context data layout

The main reason for the change was to reduce cycles in the token
decoder. (~1.5% gain for 32 bit)  This layout should be more
cache friendly.

As a result of this change, the encoder had to be updated.

Change-Id: Id5e804169d8889da0378b3a519ac04dabd28c837
Note: dixie uses a similar layout

13 years agoMerge "Fix harmless off-by-1 error."
John Koleszar [Mon, 30 Aug 2010 19:40:42 +0000 (12:40 -0700)]
Merge "Fix harmless off-by-1 error."

13 years agoMerge "Fix two-pass framrate for Y4M input."
John Koleszar [Mon, 30 Aug 2010 19:40:37 +0000 (12:40 -0700)]
Merge "Fix two-pass framrate for Y4M input."

13 years agoMerge "increase rate control buffer level precision"
John Koleszar [Mon, 30 Aug 2010 14:49:35 +0000 (07:49 -0700)]
Merge "increase rate control buffer level precision"

13 years agoFix harmless off-by-1 error.
Timothy B. Terriberry [Wed, 5 May 2010 23:14:36 +0000 (19:14 -0400)]
Fix harmless off-by-1 error.

The memory being zeroed in vp8_update_mode_info_border() was just
 allocated with calloc, and so the entire function is actually
 redundant, but it should be made correct in case someone expects
 it to actually work in the future.

Change-Id: If7a84e489157ab34ab77ec6e2fe034fb71cf8c79

13 years agoFix two-pass framrate for Y4M input.
Timothy B. Terriberry [Fri, 27 Aug 2010 22:21:22 +0000 (15:21 -0700)]
Fix two-pass framrate for Y4M input.

The timebase was being set to the value in the Y4M file on each
 pass, but only doubled to account for the altref placement on
 the first past.
This avoids reseting it on the second pass.

Change-Id: Ie342639bad1ffe9c2214fbbaaded72cfed835b42

13 years agoMerge "Allow --cpu= to work for x86."
Fritz Koenig [Wed, 25 Aug 2010 18:39:59 +0000 (11:39 -0700)]
Merge "Allow --cpu= to work for x86."

13 years agoAllow --cpu= to work for x86.
Fritz Koenig [Tue, 24 Aug 2010 23:27:49 +0000 (16:27 -0700)]
Allow --cpu= to work for x86.

--cpu was already implemented for most of our embedded
platforms, this just extends it to x86.  Corner case for
Atom processor as it doesn't respond to the --march=
option under icc.

Change-Id: I2d57a7a6e9d0b55c0059e9bc46cfc9bf9468c185

13 years agoclean up compiler warnings
Johann [Tue, 24 Aug 2010 22:23:16 +0000 (18:23 -0400)]
clean up compiler warnings

did a test compile with clang and got rid of some warnings that have
been annoying me for a while:
vp8/decoder/detokenize.c: In function 'vp8_init_detokenizer':
vp8/decoder/detokenize.c:121: warning: assignment discards qualifiers from pointer target type
vp8/decoder/detokenize.c:122: warning: assignment discards qualifiers from pointer target type
vp8/decoder/detokenize.c:123: warning: assignment from incompatible pointer type
vp8/decoder/detokenize.c:124: warning: assignment discards qualifiers from pointer target type
vp8/decoder/detokenize.c:125: warning: assignment discards qualifiers from pointer target type
vp8/decoder/detokenize.c:128: warning: assignment discards qualifiers from pointer target type
vp8/decoder/detokenize.c:129: warning: assignment discards qualifiers from pointer target type
vp8/decoder/detokenize.c:130: warning: assignment discards qualifiers from pointer target type
vp8/decoder/detokenize.c:131: warning: assignment discards qualifiers from pointer target type

Change-Id: I78ddab176fe47cbeed30379709dc7bab01c0c2e4

13 years agoupdate structures
Johann [Mon, 23 Aug 2010 17:35:26 +0000 (13:35 -0400)]
update structures

mbmi and eob moved in previous commits

Change-Id: I30a2eba36addf89ee50b406ad4afdd059a832711

13 years agoRework idct calling structure.
Fritz Koenig [Fri, 20 Aug 2010 17:58:19 +0000 (10:58 -0700)]
Rework idct calling structure.

Moving the eob structure allows for a non-struct based
function to handle decoding an entire mb of
idct/dequant/recon data.  This allows for SIMD functions
to idct/dequant/recon multiple blocks at once.

SSE2 implementation gives 3% gain on Atom.

Change-Id: I8a8f3efd546ea4e0535f517d94f347cfb737c9c2

13 years agoincrease rate control buffer level precision
John Koleszar [Fri, 20 Aug 2010 15:04:10 +0000 (11:04 -0400)]
increase rate control buffer level precision

The external API exposes the RC initial/optimal/full buffer level in
milliseconds, but this value was truncated internally to seconds. This
patch allows the use of the full precision during the conversion from
time to bits.

Change-Id: If8dd2a87614c05747f81432cbe75dd9e6ed2f04e

13 years agoRevert "Removed ssse3 sixtap code"
Jim Bankoski [Thu, 19 Aug 2010 19:50:29 +0000 (15:50 -0400)]
Revert "Removed ssse3 sixtap code"

This reverts commit 6ea5bb85cd1547b846f4c794e8684de5abcf9f62.

13 years agocleanup simple loop filter
Johann [Thu, 19 Aug 2010 17:37:40 +0000 (13:37 -0400)]
cleanup simple loop filter

move some things around, reorder some instructions

constant 0 is used several times. load it once per call in horiz,
once per loop in vert.

separate saturating instructions to avoid stalls.

just use one usub8 call to set GE flags, rather than uqsub8 followed by
usub8 w/ 0

document some stalls for further consideration

Change-Id: Ic3877e0ddbe314bb8a17fd5db73501a7d64570ec

13 years agoMerge "fix armv6 simpleloop filter"
Johann [Thu, 19 Aug 2010 15:31:57 +0000 (08:31 -0700)]
Merge "fix armv6 simpleloop filter"

13 years agofix armv6 simpleloop filter
Johann [Thu, 19 Aug 2010 15:29:21 +0000 (11:29 -0400)]
fix armv6 simpleloop filter

test cases were causing a crash because the count was being read
incorrectly. after fixing that, noticed that the output was not
matching. fixed that.

Change-Id: Idb0edb887736bd566a3cf6d4aa1a03ea8d20eb27

13 years agoRemoved ssse3 sixtap code
Scott LaVarnway [Wed, 18 Aug 2010 19:29:38 +0000 (15:29 -0400)]
Removed ssse3 sixtap code

Change-Id: I0f20fbb898ee31eb94a143471aa6f1ca17a229a4

13 years agoMerge "store more vars than we removed"
John Koleszar [Mon, 16 Aug 2010 14:54:48 +0000 (07:54 -0700)]
Merge "store more vars than we removed"

13 years agostore more vars than we removed
Johann [Mon, 16 Aug 2010 14:32:15 +0000 (10:32 -0400)]
store more vars than we removed

only saved r4-11+lr, but were storing r4-r12+lr

Change-Id: If77df1998af50e9badee7d99ef53543046434675

13 years agoarm: fix missing dependency with --enable-shared
John Koleszar [Mon, 16 Aug 2010 13:34:30 +0000 (09:34 -0400)]
arm: fix missing dependency with --enable-shared

The C version of the dequant/idct/add function depends on the C
version of the IDCT, but this isn't compiled in on ARM. Since this
code has asm version, we can just remove this file to eliminate the
link error.

Change-Id: I21de74d89d3765a1db2da27292b20727c53178e9

13 years agomove segmentation_common to encoder
John Koleszar [Fri, 13 Aug 2010 18:50:51 +0000 (14:50 -0400)]
move segmentation_common to encoder

vp8_update_gf_useage_maps() is only used by the encoder. This patch
fixes the ability to build in decode-only or encode-only
configurations.

Change-Id: I3a5211428e539886ba998e09e8abd747ac55c9aa

13 years agoframework for assembly version of the detokenizer
Johann [Thu, 12 Aug 2010 13:05:37 +0000 (09:05 -0400)]
framework for assembly version of the detokenizer

adds a compile time option: --enable-arm-asm-detok which pulls in
vp8/decoder/arm/detokenize.asm

currently about break even speed wise, but changes are pending to
the fill code (branch and load 3 bytes versus conditionally always
load one) and the error handling. Currently it doesn't handle zero
runs or overrunning the buffer.

this is really just so i don't have to rebase my changes all the
time to run benchmarks - now just need to replace one file!

Change-Id: I56d0e2354dc0ca3811bffd0e88fe1f952fa6c797

13 years agoupdate structure
Johann [Thu, 12 Aug 2010 17:27:07 +0000 (13:27 -0400)]
update structure

mode_info_context->mbmi no longer gets copied up a level

Change-Id: Icd2d27d381909721326c34594a1ccdc26d48a995

13 years agoremove unused definition
Johann [Thu, 12 Aug 2010 17:06:47 +0000 (13:06 -0400)]
remove unused definition

asm_offsets contains some definitions which are no longer used. this
was one of them. v6 build works now

Change-Id: If370cfa8acd145de4fead2d9a11b048fccc090df

13 years agoRemoved unnecessary MB_MODE_INFO copies
Scott LaVarnway [Thu, 12 Aug 2010 20:25:43 +0000 (16:25 -0400)]
Removed unnecessary MB_MODE_INFO copies

These copies occurred for each macroblock in the encoder and decoder.
Thetemp MB_MODE_INFO mbmi was removed from MACROBLOCKD.  As a result,
a large number compile errors had to be fixed.

Change-Id: I4cf0ffae3ce244f6db04a4c217d52dd256382cf3

13 years agoMerge "Finished vp8_sixtap_predict4x4_ssse3 function"
Scott LaVarnway [Wed, 11 Aug 2010 19:23:24 +0000 (12:23 -0700)]
Merge "Finished vp8_sixtap_predict4x4_ssse3 function"

13 years agocosmetics: add missing 2D array braces
John Koleszar [Mon, 9 Aug 2010 17:48:04 +0000 (13:48 -0400)]
cosmetics: add missing 2D array braces

Silences compile warning.

Change-Id: I4b207d97f8570fe29aa2710e4ce4f02e7e43b57a

13 years agoavoid negative array subscript warnings
John Koleszar [Mon, 9 Aug 2010 17:27:26 +0000 (13:27 -0400)]
avoid negative array subscript warnings

The mv_ref and sub_mv_ref token encodings are indexed from NEARESTMV
and LEFT4X4, respectively, rather than being zero-based like the
other token encodings.

Change-Id: I3699c3f84111209ecfb91097c4b900773e9a3ad5

13 years agoFinished vp8_sixtap_predict4x4_ssse3 function
Scott LaVarnway [Wed, 11 Aug 2010 17:49:00 +0000 (13:49 -0400)]
Finished vp8_sixtap_predict4x4_ssse3 function

Added vp8_filter_block1d4_h6_ssse3 and vp8_filter_block1d4_v6_ssse3
assembly routines.  Also removed unused assembly.

Change-Id: I01c1021835f2edda9da706822345f217087ca0d0

13 years agorename DETOK_[AL]
Johann [Wed, 11 Aug 2010 17:36:35 +0000 (13:36 -0400)]
rename DETOK_[AL]

everything else uses lowercase detok

Change-Id: I9671e2e90eb2961208dfa81c00b3accb5749ec04

13 years agoMoved gf_active code to encoder only
Scott LaVarnway [Wed, 11 Aug 2010 15:02:31 +0000 (11:02 -0400)]
Moved gf_active code to encoder only

The gf_active code is only used by the encoder, so it was moved from
common and decoder.

Change-Id: Iada15acd5b2b33ff70c34668ca87d4cfd0d05025

13 years agoRemoved duplicate functions
Yaowu Xu [Wed, 11 Aug 2010 04:45:34 +0000 (21:45 -0700)]
Removed duplicate functions

Change-Id: Ie587972ccefd3c762b8cdf8ef39345cd22924b9b

13 years agoNormalize quantizer's zero bin and rounding factors
Yaowu Xu [Wed, 11 Aug 2010 04:12:04 +0000 (21:12 -0700)]
Normalize quantizer's zero bin and rounding factors

This patch changes a few numbers in the two constant arrays
for quantizer's zerobin and rounding factors, in general to
make the sum of the two factors for any Q to be 128.  While
it might be beneficial to calibrate the two arrays for best
quantizer performance, it is not the purpose of this patch.
Normalizing the two arrays will enable quick optimization
of the current faster quantizer, i.e .zerobin check can be
removed.

Change-Id: If9abfd7929bf4b8e9ecd64a79d817c6728c820bd

13 years agoAdd trellis quantization.
Timothy B. Terriberry [Fri, 2 Jul 2010 21:35:53 +0000 (14:35 -0700)]
Add trellis quantization.

Replace the exponential search for optimal rounding during
 quantization with a linear Viterbi trellis and enable it
 by default when using --best.
Right now this operates on top of the output of the adaptive
 zero-bin quantizer in vp8_regular_quantize_b() and gives a small
 gain.
It can be tested as a replacement for that quantizer by
 enabling the call to vp8_strict_quantize_b(), which uses
 normal rounding and no zero bin offset.
Ultimately, the quantizer will have to become a function of lambda
 in order to take advantage of activity masking, since there is
 limited ability to change the quantization factor itself.
However, currently vp8_strict_quantize_b() plus the trellis
 quantizer (which is lambda-dependent) loses to
 vp8_regular_quantize_b() alone (which is not) on my test clip.

Patch Set 3:

Fix an issue related to the cost evaluation of successor
states when a coefficient is reduced to zero. With this
issue fixed, now the trellis search almost exactly matches
the exponential search.

Patch Set 2:

Overall, the goal of this patch set is to make "trellis"
search to produce encodings that match the exponential
search version. There are three main differences between
Patch Set 2 and 1:
a. Patch set 1 did not properly account for the scale of
2nd order error, so patch set 2 disable it all together
for 2nd blocks.
b. Patch set 1 was not consistent on when to enable the
the quantization optimization. Patch set 2 restore the
condition to be consistent.
c. Patch set 1 checks quantized level L-1, and L for any
input coefficient was quantized to L. Patch set 2 limits
the candidate coefficient to those that were rounded up
to L. It is worth noting here that a strategy to check
L and L+1 for coefficients that were truncated down to L
might work.

(a and b get trellis quant to basically match the exponential
search on all mid/low rate encodings on cif set, without
a, b, trellis quant can hurt the psnr by 0.2 to .3db at
200kbps for some cif clips)
(c gets trellis quant  to match the exponential search
to match at Q0 encoding, without c, trellis quant can be
1.5 to 2db lower for encodings with fixed Q at 0 on most
derf cif clips)

Change-Id: Ib1a043b665d75fbf00cb0257b7c18e90eebab95e

13 years agoAdded ssse3 version of sixtap filters
Scott LaVarnway [Tue, 10 Aug 2010 21:06:05 +0000 (17:06 -0400)]
Added ssse3 version of sixtap filters

Improved decoder performance by 9% for the clip used.

Change-Id: I8fc5609213b7bef10248372595dc85b29f9895b9

13 years agoFirst modification of multi-thread decoder
Yunqing Wang [Thu, 29 Jul 2010 20:24:26 +0000 (16:24 -0400)]
First modification of multi-thread decoder

This is the first modification of VP8 multi-thread decoder, which uses
same threads to decode macroblocks and then do loopfiltering for each
frame.

Inspired by Rob Clark, synchronization was done on every 8 macroblocks
instead of every macroblock to reduce lock contention.

Comparing with the original code, this implementation gave about 15%-
20% performance gain while decoding my test clips on a Core2 Quad
platform (Linux).

The work is not done yet.

Test on other platforms are needed.

Change-Id: Ice9ddb0b511af1359b9f71e65066143c04fef3b5

13 years agoMark loopfilter C functions as static
John Koleszar [Mon, 9 Aug 2010 13:33:00 +0000 (09:33 -0400)]
Mark loopfilter C functions as static

Clang defaults to C99 mode, and inline works differently in C99.
(gcc, on the other hand, defaults to a special gnu-style inlining,
which uses different syntax.)   Making the functions static makes sure
clang doesn't decide to discard a function because it's too large to
inline.

Thanks to eli.friedman for the patch.

Fixes http://code.google.com/p/webm/issues/detail?id=114

Change-Id: If3c1c3c176eb855a584a60007237283b0cc631a4

13 years agoMerge "Issue 150: Fixing linker warning in extend.c."
John Koleszar [Mon, 2 Aug 2010 16:35:05 +0000 (09:35 -0700)]
Merge "Issue 150: Fixing linker warning in extend.c."

13 years agoconfigure: support directories containing .o
John Koleszar [Mon, 2 Aug 2010 14:21:52 +0000 (10:21 -0400)]
configure: support directories containing .o

Fixes http://code.google.com/p/webm/issues/detail?id=96

The regex which postprocesses the gcc make-deps (-M) output was too
greedy and matching in the dependencies part of the rule rather than
the target only. The patch provided with the issue was not correct, as
it tried to match the .o at the end of the line, which isn't correct
at least for my GCC version. This patch matches word characters
instead of .*

Thanks to raimue and the MacPorts community for isolating this issue.

Change-Id: I28510da2252e03db910c017101d9db12e5945a27

13 years agonasm: avoid space before the :data symbol type.
Jan Kratochvil [Sat, 31 Jul 2010 15:12:31 +0000 (17:12 +0200)]
nasm: avoid space before the :data symbol type.

global label:data
           ^^

Provide nasm compatibility.  No binary change by this patch with yasm
on {x86_64,i686}-fedora13-linux-gnu.  Few longer opcodes with nasm on
{x86_64,i686}-fedora13-linux-gnu have been checked as safe.

Change-Id: I10f17eb1e4d4a718d4ebd1d0ccddc807c365e021

13 years agonasm: end labels with colon (':')
Jan Kratochvil [Sat, 31 Jul 2010 15:12:31 +0000 (17:12 +0200)]
nasm: end labels with colon (':')

Labels should end by colon (':'), nasm requires it.

Provide nasm compatibility.  No binary change by this patch with yasm
on {x86_64,i686}-fedora13-linux-gnu.  Few longer opcodes with nasm on
{x86_64,i686}-fedora13-linux-gnu have been checked as safe.

Change-Id: I0b2ec6f01afb061d92841887affb5ca0084f936f

13 years agonasm: use OWORD vs DQWORD
Jan Kratochvil [Sat, 31 Jul 2010 15:12:31 +0000 (17:12 +0200)]
nasm: use OWORD vs DQWORD

nasm knows only OWORD.  yasm knows both OWORD and DQWORD.

Provide nasm compatibility.  No binary change by this patch with yasm on
{x86_64,i686}-fedora13-linux-gnu.  Few longer opcodes with nasm on
{x86_64,i686}-fedora13-linux-gnu have been checked as safe.

Change-Id: I62151390089e90df9a7667822fa594ac20b00e78

13 years agoMerge "Replace pinsrw (SSE) with MMX instructions"
John Koleszar [Mon, 2 Aug 2010 13:16:26 +0000 (06:16 -0700)]
Merge "Replace pinsrw (SSE) with MMX instructions"

13 years agoReplace pinsrw (SSE) with MMX instructions
Philip Jägenstedt [Tue, 13 Jul 2010 09:43:51 +0000 (11:43 +0200)]
Replace pinsrw (SSE) with MMX instructions

Fixes http://code.google.com/p/webm/issues/detail?id=136

Change-Id: I5a3e294061644a1a9718e8ba4a39548ede25cc42

13 years agoapple: include proper mach primatives
John Koleszar [Thu, 29 Jul 2010 21:04:39 +0000 (17:04 -0400)]
apple: include proper mach primatives

Fixes implicit declaration warning for 'mach_task_self'.

Patch courtesy of timeless at gmail.com

Change-Id: I9991dedd1ccfddc092eca86705ecbc3b764b799d

13 years agoMerge "Enable the switch between two versions of quantizer"
Yaowu Xu [Thu, 29 Jul 2010 14:17:40 +0000 (07:17 -0700)]
Merge "Enable the switch between two versions of quantizer"

13 years agoRemoved two unused global variables.
Frank Galligan [Wed, 28 Jul 2010 21:25:09 +0000 (17:25 -0400)]
Removed two unused global variables.

Removed the global variables vp8_an and vp8_cd. vp8_an was causing problems
because it was increasing the .bss by 1572864 bytes.

Change-Id: I6c12e294133c7fb6e770c0e4536d8287a5720a87

13 years agoEnable the switch between two versions of quantizer
Yaowu Xu [Wed, 28 Jul 2010 17:44:17 +0000 (10:44 -0700)]
Enable the switch between two versions of quantizer

To facilitate more testing related to quantizer and rate
control, the old version quantizer is added back. old and
new quantizer can be switched back and forth by define or
un-define the macro "EXACT_QUANT".

Change-Id: Ia77e687622421550f10e9d65a9884128a79a65ff

13 years agoconfigure: pass original arguments through to make dist
John Koleszar [Tue, 22 Jun 2010 13:53:23 +0000 (09:53 -0400)]
configure: pass original arguments through to make dist

When running configure automatically through the make dist target,
reuse the arguments passed to the original configure command.

Change-Id: I40e5b8384d6485a565b91e6d2356d5bc9c4c5928

13 years agoMerge "msvs: fix install of codec sources"
John Koleszar [Tue, 27 Jul 2010 18:21:42 +0000 (11:21 -0700)]
Merge "msvs: fix install of codec sources"

13 years agox86/sse2: disable asm quantizer
Johann [Tue, 27 Jul 2010 16:10:48 +0000 (12:10 -0400)]
x86/sse2: disable asm quantizer

follow up to Change I0e51492d: neon: disable asm quantizer

Now x86 doesn't segfault with --disable-runtime-cpu-detect and -p=2

Change-Id: I8ca127bb299198efebbcbd5a661e81788361933f

13 years agoFix build w/o RTCD
Johann [Tue, 27 Jul 2010 15:56:19 +0000 (11:56 -0400)]
Fix build w/o RTCD

So many places to update ...

Change-Id: Ide957b40cc833f99c2d1849acade6850fbf7585d

13 years agoneon: disable asm quantizer
John Koleszar [Tue, 27 Jul 2010 15:12:21 +0000 (11:12 -0400)]
neon: disable asm quantizer

The assembly version of the quantizer has not been updated to match
the new exact quantizer introduced in commit e04e2935. That commit tried
to disable this code but missed the non-RTCD case.

Thanks to David Baker <david.baker at openmarket.com> for isolating the
issue and testing this fix.

Change-Id: I0e51492dc6f8e44d2c10b587427448bf94135c65

13 years agoMerge "update arm idct functions"
Fritz Koenig [Mon, 26 Jul 2010 13:05:39 +0000 (06:05 -0700)]
Merge "update arm idct functions"

13 years agoMerge changes I896fe6f9,I90d8b167
Fritz Koenig [Mon, 26 Jul 2010 13:05:30 +0000 (06:05 -0700)]
Merge changes I896fe6f9,I90d8b167

* changes:
  Change the x86 idct functions to do reconstruction at the same time
  Combine idct and reconstruction steps

13 years agoupdate arm idct functions
Johann [Fri, 23 Jul 2010 17:42:30 +0000 (13:42 -0400)]
update arm idct functions

Jeff Muizelaar posted some changes to the idct/reconstruction c code.
This is the equivalent update for the arm assembly.

This shows a good boost on v6, and a minor boost on neon.
Here are some numbers for highway in qcif, 2641 frames:
HEAD neon: ~161 fps
new neon:  ~162 fps
HEAD v6:   ~102 fps
new v6:    ~106 fps

The following functions have been updated for armv6 and neon:
vp8_dc_only_idct_add
vp8_dequant_idct_add
vp8_dequant_dc_idct_add

Conflicts:

vp8/decoder/arm/armv6/dequantdcidct_v6.asm
vp8/decoder/arm/armv6/dequantidct_v6.asm

Resolved by removing these files. When I rewrote the functions, I also
moved the files to dequant_dc_idct_v6.asm/dequant_idct_v6.asm

Change-Id: Ie3300df824d52474eca1a5134cf22d8b7809a5d4

13 years agoIssue 150: Fixing linker warning in extend.c.
Justin Lebar [Fri, 23 Jul 2010 23:42:25 +0000 (16:42 -0700)]
Issue 150: Fixing linker warning in extend.c.

13 years agoDon't dereference ctx->priv if it hasn't been setup correctly.
Fredrik Söderquist [Mon, 7 Jun 2010 16:24:41 +0000 (18:24 +0200)]
Don't dereference ctx->priv if it hasn't been setup correctly.

13 years agoOnly touch ctx->priv if vp8_mmap_alloc succeeded.
Fredrik Söderquist [Mon, 7 Jun 2010 16:20:47 +0000 (18:20 +0200)]
Only touch ctx->priv if vp8_mmap_alloc succeeded.

13 years agoChange the x86 idct functions to do reconstruction at the same time
Jeff Muizelaar [Thu, 3 Jun 2010 14:16:07 +0000 (10:16 -0400)]
Change the x86 idct functions to do reconstruction at the same time

Change-Id: I896fe6f9664e6849c7cee2cc6bb4e045eb42540f

13 years agoCombine idct and reconstruction steps
Jeff Muizelaar [Fri, 28 May 2010 18:28:12 +0000 (14:28 -0400)]
Combine idct and reconstruction steps

This moves the prediction step before the idct and combines the idct and
reconstruction steps into a single step. Combining them seems to give an
overall decoder performance improvement of about 1%.

Change-Id: I90d8b167ec70d79c7ba2ee484106a78b3d16e318

13 years agoSwap alt/gold/new/last frame buffer ptrs instead of copying.
Fritz Koenig [Thu, 22 Jul 2010 12:07:32 +0000 (08:07 -0400)]
Swap alt/gold/new/last frame buffer ptrs instead of copying.

At the end of the decode, frame buffers were being copied.
The frames are not updated after the copy, they are just
for reference on later frames.  This change allows multiple
references to the same frame buffer instead of copying it.

Changes needed to be made to the encoder to handle this.  The
encoder is still doing frame buffer copies in similar places
where pointer reference could be done.

Change-Id: I7c38be4d23979cc49b5f17241ca3a78703803e66

13 years agoMerge commit 'refs/changes/51/351/1' of ssh://review.webmproject.org:29418/libvpx...
Paul Wilkins [Fri, 23 Jul 2010 16:45:26 +0000 (17:45 +0100)]
Merge commit 'refs/changes/51/351/1' of ssh://review.webmproject.org:29418/libvpx into KfRateBugMerged

13 years agoMerge "Make the quantizer exact."
Yaowu Xu [Fri, 23 Jul 2010 16:26:26 +0000 (09:26 -0700)]
Merge "Make the quantizer exact."

13 years agoRate control bug with long key frame interval.
Paul Wilkins [Fri, 23 Jul 2010 16:01:12 +0000 (17:01 +0100)]
Rate control bug with long key frame interval.

In two pass encodes, the calculation of the number of bits
allocated to a KF group had the potential to overflow for high data
rates if the interval is very long.

We observed the problem in one test clip where there was one
section where there was an 8000 frame gap between key frames.

Change-Id: Ic48eb86271775d7573b4afd166b567b64f25b787

13 years agoMake the quantizer exact.
Timothy B. Terriberry [Tue, 29 Jun 2010 00:15:09 +0000 (17:15 -0700)]
Make the quantizer exact.

This replaces the approximate division-by-multiplication in the
 quantizer with an exact one that costs just one add and one
 shift extra.
The asm versions have not been updated in this patch, and thus
 have been disabled, since the new method requires different
 multipliers which are not compatible with the old method.

Change-Id: I53ac887af0f969d906e464c88b1f4be69c6b1206

13 years ago80 character line length on Arnr LUT
Paul Wilkins [Fri, 23 Jul 2010 15:47:54 +0000 (16:47 +0100)]
80 character line length on Arnr LUT

Tweaked table to fit to 80 characters.

Change-Id: Ie6ba80e0b31b33e23d2bf78599abe223369fcefb

13 years agoRemove CONFIG_NEW_TOKENS files.
Fritz Koenig [Thu, 22 Jul 2010 13:46:54 +0000 (09:46 -0400)]
Remove CONFIG_NEW_TOKENS files.

These files were out of date and no longer maintained.
Token decoding has implemented the no-crash code which
is incompatible with this arm assembly code.

Change-Id: Ibf729886c56fca48181af60b44bda896c30023fc

13 years agomsvs: fix install of codec sources
John Koleszar [Tue, 22 Jun 2010 12:44:48 +0000 (08:44 -0400)]
msvs: fix install of codec sources

The libs.mk file must be installed for the vpx.vcproj file to be
generated. It was being installed, but not in the src/ directory as
expected.

Also missed include files yasm.rules, quantize_x86.h

Change-Id: Ic1a6f836e953bfc954d6e42a18c102a0114821eb

13 years agoChange devenv.com command line.
Tom Finegan [Thu, 22 Jul 2010 21:51:17 +0000 (17:51 -0400)]
Change devenv.com command line.

Change /build to -build to avoid problems when builds are run within
msys bash shells.

Change-Id: Ie68d72f702adad00d99be8a01c7a388c3af7657d

13 years agoAdd vs9 targets.
Tom Finegan [Thu, 22 Jul 2010 17:34:25 +0000 (13:34 -0400)]
Add vs9 targets.

Add targets x86-win32-vs9 and x86_64-win64-vs9 for support of Visual
Studio 2008-- this removes the need to convert the vs8 projects before
using them within the IDE.

Change-Id: Idb83e2ae701e07d98db1be71638280a493d770a2

13 years agoMerge "limit range checking code for L[k] to CONFIG_DEBUG. patch by timeless@gmail...
Johann [Wed, 21 Jul 2010 19:59:39 +0000 (12:59 -0700)]
Merge "limit range checking code for L[k] to CONFIG_DEBUG. patch by timeless@gmail.com"

13 years agoMerge "Improve the accuracy of forward walsh-hadamard transform"
Yaowu Xu [Mon, 19 Jul 2010 14:50:26 +0000 (07:50 -0700)]
Merge "Improve the accuracy of forward walsh-hadamard transform"

13 years agoARNR Lookup Table.
Paul Wilkins [Mon, 19 Jul 2010 12:28:34 +0000 (13:28 +0100)]
ARNR Lookup Table.

Change submitted for Adrian Grange. Convert threshold
calculation in ARNR filter to a lookup table.

Change-Id: I12a4bbb96b9ce6231ce2a6ecc2d295610d49e7ec

13 years agoParameter limit change.
Paul Wilkins [Mon, 19 Jul 2010 10:32:09 +0000 (11:32 +0100)]
Parameter limit change.

Change maximum ARNR filter width to 15.

Change-Id: I3b72450ea08e96287445ec18810630ee2292954c

13 years agoRate control fix for ARNR filtered frames.
Paul Wilkins [Mon, 19 Jul 2010 13:10:07 +0000 (14:10 +0100)]
Rate control fix for ARNR filtered frames.

Previously we had assumed that it was necessary to give a full frame's
bit allocation to the alt ref frame if it has been created through temporal
filtering. This is not the case. The active max quantizer control
insures that sufficient bits are allocated if needed and allocating a
full frame's worth of bits creates an excessive overhead for the ARF.

Change-Id: I83c95ed7bc7ce0e53ccae6ff32db5a97f145937a

13 years agoFix: Incorrect 'cols' calculation in temporal filter.
Paul Wilkins [Fri, 16 Jul 2010 14:57:17 +0000 (15:57 +0100)]
Fix: Incorrect 'cols' calculation in temporal filter.

Change-Id: I37f10fbe4fbb505c1d34980a59af3e817c287e22

13 years agolimit range checking code for L[k] to CONFIG_DEBUG. patch by timeless@gmail.com
Michael Kohler [Mon, 12 Jul 2010 16:41:45 +0000 (18:41 +0200)]
limit range checking code for L[k] to CONFIG_DEBUG. patch by timeless@gmail.com

13 years agomsvs: disable CRT deprecation warnings
John Koleszar [Tue, 22 Jun 2010 13:45:43 +0000 (09:45 -0400)]
msvs: disable CRT deprecation warnings

Disables the warnings produced for so-called insecure standard C
functions.

Change-Id: I0e6f448e27f899a0eaefc1151185945fbe15718e

13 years agoMerge "Fix misspelled "paramter" in vpx_codec_internal.h" to "parameter"."
John Koleszar [Wed, 7 Jul 2010 23:57:37 +0000 (16:57 -0700)]
Merge "Fix misspelled "paramter" in vpx_codec_internal.h" to "parameter"."

13 years agoMerge "Fix misspelled "skiped" in onyxc_int.h to "skipped"."
John Koleszar [Wed, 7 Jul 2010 23:57:08 +0000 (16:57 -0700)]
Merge "Fix misspelled "skiped" in onyxc_int.h to "skipped"."

13 years agoFix a compiling error on armv6
Yaowu Xu [Wed, 7 Jul 2010 17:26:30 +0000 (10:26 -0700)]
Fix a compiling error on armv6

The issue was caused by a bad merge in Change I5559d1e8

Change-Id: I6563f652bc1500202de361f8f51d11cc6ddf3331

13 years agoFix misspelled "paramter" in vpx_codec_internal.h" to "parameter".
Michael Kohler [Wed, 7 Jul 2010 17:49:58 +0000 (19:49 +0200)]
Fix misspelled "paramter" in vpx_codec_internal.h" to "parameter".

Signed-off-by: Michael Kohler <michaelkohler@live.com>
13 years agoFix misspelled "skiped" in onyxc_int.h to "skipped".
Michael Kohler [Wed, 7 Jul 2010 17:48:12 +0000 (19:48 +0200)]
Fix misspelled "skiped" in onyxc_int.h to "skipped".

Signed-off-by: Michael Kohler <michaelkohler@live.com>
13 years agoFix bug in 1st pass motion compensation
Adrian Grange [Thu, 1 Jul 2010 13:17:04 +0000 (14:17 +0100)]
Fix bug in 1st pass motion compensation

In the case where the best reference mv is not (0,0) a secondary
search is carried out centered on (0,0). However, rather than
sending tmp_err into the search function, motion_error was
inadvertently passed.

As a result tmp_err remains set at INT_MAX and the (0,0)-based
search result will never be selected, even if it is better.

Change-Id: I3c82b246c8c82ba887b9d3fb4c9e0a0f2fe5a76c

13 years agoMerge "Further adjustment of RD behaviour with Q and Zbin."
Paul Wilkins [Thu, 1 Jul 2010 08:53:40 +0000 (01:53 -0700)]
Merge "Further adjustment of RD behaviour with Q and Zbin."

13 years agoMerge "ivfenc: correct fixed kf interval, --disable-kf"
John Koleszar [Thu, 1 Jul 2010 00:19:26 +0000 (17:19 -0700)]
Merge "ivfenc: correct fixed kf interval, --disable-kf"

13 years agoMerge "ARM WinCE VS8 build update"
John Koleszar [Wed, 30 Jun 2010 15:01:02 +0000 (08:01 -0700)]
Merge "ARM WinCE VS8 build update"

13 years agoMerge "Remove INLINE/FORCEINLINE"
John Koleszar [Wed, 30 Jun 2010 14:59:39 +0000 (07:59 -0700)]
Merge "Remove INLINE/FORCEINLINE"

13 years agoUpdate loopfilter frame/filter/sharp info for multithread
John Koleszar [Wed, 30 Jun 2010 14:22:40 +0000 (10:22 -0400)]
Update loopfilter frame/filter/sharp info for multithread

Change I9fd1a5a4 updated the multithreaded loopfilter to avoid
reinitializing several parameteres if they haven't changed from the
last frame, but the code to update the last frame's parameters wasn't
invoked in the multithreaded case.

Change-Id: Ia23d937af625c01dd739608e02d110f742b7e1f2

13 years agoMerge "Add loopfilter initialization fix in multithreading code"
Yunqing Wang [Wed, 30 Jun 2010 13:56:36 +0000 (06:56 -0700)]
Merge "Add loopfilter initialization fix in multithreading code"

13 years agoAdd loopfilter initialization fix in multithreading code
Yunqing Wang [Wed, 30 Jun 2010 13:42:39 +0000 (09:42 -0400)]
Add loopfilter initialization fix in multithreading code

Modified loopfilter initialization to avoid unnecessary operations.

Change-Id: I9fd1a5a49edc1cb8116c2a72a6908b1e437459ec

13 years agoMerge "Fixed buffer selection for UV in AltRef filtering"
Adrian Grange [Wed, 30 Jun 2010 09:43:47 +0000 (02:43 -0700)]
Merge "Fixed buffer selection for UV in AltRef filtering"

13 years agoImprove SSE2 loopfilter functions
Yunqing Wang [Fri, 25 Jun 2010 13:18:11 +0000 (09:18 -0400)]
Improve SSE2 loopfilter functions

Restructured and rewrote SSE2 loopfilter functions. Combined u and
v into one function to take advantage of SSE2 128-bit registers.
Tests on test clips showed a 4% decoder performance improvement on
Linux desktop.

Change-Id: Iccc6669f09e17f2224da715f7547d6f93b0a4987

13 years agoARM WinCE VS8 build update
James Zern [Tue, 29 Jun 2010 16:02:19 +0000 (12:02 -0400)]
ARM WinCE VS8 build update

The generated project is vpx.vcproj, change vpx_decoder references to
match. Remove .rules file dependency as it will be pulled from the
source tree.

Change-Id: I679db2748b37adae3bafd764dba8575fc3abde72

13 years agoFurther adjustment of RD behaviour with Q and Zbin.
Paul Wilkins [Tue, 29 Jun 2010 11:15:54 +0000 (12:15 +0100)]
Further adjustment of RD behaviour with Q and Zbin.

Following conversations with Tim T (Derf) I ran a large number of
tests comparing the existing polynomial expression with a simpler
^2 variant. Though the polynomial was sometimes a little better at
the extremes of Q it was possible to get close for most clips and
even a little better on some.

This code also changes the way the RD multiplier is calculated
when the ZBIN is extended to use a variant of the same ^2
expression.

I hope that this simpler expression will be easier to tune further
as we expand our test set and consider adjustments based on content.

Change-Id: I73b2564346e74d1332c33e2c1964ae093437456c

13 years agoImprove the accuracy of forward walsh-hadamard transform
Yaowu Xu [Tue, 29 Jun 2010 05:03:43 +0000 (22:03 -0700)]
Improve the accuracy of forward walsh-hadamard transform

Besides the slight improvement in round trip error. This
also fixes a sign bias in the forward transform, so the
round trip errors are evenly distributed between +1s and
-1s. The old bias seemed to work well with the dc sign bias
in old fdct,  which no longer exist in the improved fdct.

Change-Id: I8635e7be16c69e69a8669eca5438550d23089cef

13 years agoFixed buffer selection for UV in AltRef filtering
Adrian Grange [Mon, 28 Jun 2010 11:00:11 +0000 (12:00 +0100)]
Fixed buffer selection for UV in AltRef filtering

Corrected setting of "which_buffer" for U & V cases to match that
used for Y, i.e. to refer to the temporally most recent frame of
those to be filtered.

Change-Id: Idf94b287ef47a05f060da3e61134a0b616adcb6b