profile/ivi/libvpx.git
13 years agoModification and issue fix in full-pixel refining search
Yunqing Wang [Wed, 11 May 2011 17:38:29 +0000 (13:38 -0400)]
Modification and issue fix in full-pixel refining search

Further modification and wrong implementation fix which caused
refining_search and refining_searchx4 result mismatching.

Change-Id: I80cb3a44bf5824413fd50c972e383eebb75f9b6f

13 years agoMerge "Use diamond search to replace full search in full-pixel refining search"
Yunqing Wang [Tue, 10 May 2011 13:59:38 +0000 (06:59 -0700)]
Merge "Use diamond search to replace full search in full-pixel refining search"

13 years agoUse diamond search to replace full search in full-pixel refining search
Yunqing Wang [Fri, 6 May 2011 16:51:31 +0000 (12:51 -0400)]
Use diamond search to replace full search in full-pixel refining search

In NEWMV mode, currently, full search is used as the refining search
after n-step search. By replacing it with an iterative diamond search
of radius 1 largely reduced the computation complexity, but still
maintained the same encoding quality since the refining search is
done for every macroblock instead of only a small precentage of
macroblocks while using full search.

Tests on the test set showed a 3.4% encoding speed increase with none
psnr & ssim loss.

Change-Id: Ife907d7eb9544d15c34f17dc6e4cfd97cb743d41

13 years agoclean up unused variable warnings
Johann [Mon, 9 May 2011 15:16:31 +0000 (11:16 -0400)]
clean up unused variable warnings

Change-Id: I9467d7a50eac32d8e8f3a2f26db818e47c93c94b

13 years agoMerge "Don't override active_worst_quality in 2 pass"
John Koleszar [Fri, 6 May 2011 15:59:05 +0000 (08:59 -0700)]
Merge "Don't override active_worst_quality in 2 pass"

13 years agoMerge "neon fast quantizer updated"
Johann [Fri, 6 May 2011 15:54:14 +0000 (08:54 -0700)]
Merge "neon fast quantizer updated"

13 years agoDon't override active_worst_quality in 2 pass
John Koleszar [Fri, 6 May 2011 15:48:50 +0000 (11:48 -0400)]
Don't override active_worst_quality in 2 pass

Commit db5057c introduced a bug in that the active_worst_quality
selected by the 2 pass rate controller was being overridden for key
frames, causing a severe quality loss.

Change-Id: I4865a6fbe3e94e9b4fb9271c7dd68b455d7b371d

13 years agoneon fast quantizer updated
Tero Rintaluoma [Mon, 11 Apr 2011 09:04:17 +0000 (12:04 +0300)]
neon fast quantizer updated

vp8_fast_quantize_b_neon function updated and further optimized.
 - match current C implementation of fast quantizer
 - updated to use asm_enc_offsets for structure members
 - updated ads2gas scripts to handle alignment issues

Change-Id: I5cbad9c460ad8ddb35d2970a8684cc620711c56d

13 years agoFix semaphore emulation on Windows
Aron Rosenberg [Fri, 6 May 2011 04:10:37 +0000 (00:10 -0400)]
Fix semaphore emulation on Windows

The existing emulation of posix semaphores on Windows uses SetEvent()
and WaitForSingleObject(), which implements a binary semaphore, not a
counting semaphore as implemented by posix. This causes deadlock when
used with the expected posix semantics. Instead, this patch uses the
CreateSemaphore() and ReleaseSemaphore() calls (introduced in Windows
2000) which have the expected behavior.

This patch also reverts commit eb16f00, which split a semaphore that
was being used with counting semantics into two binary semaphores.
That commit is unnecessary with corrected emulation.

Change-Id: If400771536a27af4b0c3a31aa4c4e9ced89ce6a0

13 years agoFix rare hang in multi-thread encoder on Windows
Yunqing Wang [Thu, 5 May 2011 14:42:29 +0000 (10:42 -0400)]
Fix rare hang in multi-thread encoder on Windows

This patch is to fix a rare hang in multi-thread encoder that was
only seen on Windows. Thanks for John's help in debugging the
problem. More test is needed.

Change-Id: Idb11c6d344c2082362a032b34c5a602a1eea62fc

13 years agoMerge "Loopfilter NEON: Use VMOV for constant vectors instead of VLD."
Johann [Thu, 5 May 2011 13:16:21 +0000 (06:16 -0700)]
Merge "Loopfilter NEON: Use VMOV for constant vectors instead of VLD."

13 years agoMerge "Runtime detection of available processor cores."
Yunqing Wang [Thu, 5 May 2011 11:59:54 +0000 (04:59 -0700)]
Merge "Runtime detection of available processor cores."

13 years agoLoopfilter NEON: Use VMOV for constant vectors instead of VLD.
Attila Nagy [Wed, 4 May 2011 07:51:26 +0000 (10:51 +0300)]
Loopfilter NEON: Use VMOV for constant vectors instead of VLD.

Change-Id: I562b6e01c32bb51d00f3b95faf757fc7dc29a3a3

13 years agoMerge "Modify HEX search"
Yunqing Wang [Tue, 3 May 2011 18:59:32 +0000 (11:59 -0700)]
Merge "Modify HEX search"

13 years agoModify HEX search
Yunqing Wang [Mon, 2 May 2011 17:21:59 +0000 (13:21 -0400)]
Modify HEX search

Changed 8-neighbor searching to 4-neighour searching, and continued
searching until the center point is the best match.

Test on test set showed 1.3% encoding speed improvement as well as
0.1% PSNR and SSIM improvement at speed=-5 (rt mode).

Will continue to improve it.

Change-Id: If4993b1907dd742b906fd3f86fee77cc5932ee9a

13 years agoMerge "change to use fast ssim code for internal ssim calculations"
Yaowu Xu [Tue, 3 May 2011 18:20:52 +0000 (11:20 -0700)]
Merge "change to use fast ssim code for internal ssim calculations"

13 years agochange to use fast ssim code for internal ssim calculations
Yaowu Xu [Mon, 2 May 2011 22:27:14 +0000 (15:27 -0700)]
change to use fast ssim code for internal ssim calculations

The commit also removed the slow ssim calculation that uses a 7x7
kernel, and revised the comments to better describe how sample ssim
values are computed and averaged

Change-Id: I1d874073cddca00f3c997f4b9a9a3db0aa212276

13 years agobuild: change LDFLAGS/CFLAGS ordering.
Ronald S. Bultje [Mon, 2 May 2011 17:56:41 +0000 (13:56 -0400)]
build: change LDFLAGS/CFLAGS ordering.

Always use CFLAGS/LDFLAGS that point to headers and libvpx.a inside our
build tree before ones from the environment, which could reference
headers or libs outside the build tree.

This fixes issue 307.

Change-Id: I34d176b8c21098f6da5ea71f0147d3c49283cc45

13 years agoMerge "Fix documentation typos"
John Koleszar [Mon, 2 May 2011 13:50:22 +0000 (06:50 -0700)]
Merge "Fix documentation typos"

13 years agoFix compile error with --enable-postproc-visualizer
John Koleszar [Mon, 2 May 2011 13:28:37 +0000 (09:28 -0400)]
Fix compile error with --enable-postproc-visualizer

Typo.

Change-Id: I9cc6a4587c3d93c9f0da5e101d376741fc9622a4

13 years agoFix documentation typos
Thijs Vermeir [Fri, 29 Apr 2011 21:29:28 +0000 (23:29 +0200)]
Fix documentation typos

Change-Id: I97124670926433bf1593c91660d8b8f8482ea9ce

13 years agoMerge "Use insertion sort instead of quick sort"
Yunqing Wang [Fri, 29 Apr 2011 15:27:58 +0000 (08:27 -0700)]
Merge "Use insertion sort instead of quick sort"

13 years agoMerge "Consolidated build inter predictors"
Scott LaVarnway [Fri, 29 Apr 2011 14:13:49 +0000 (07:13 -0700)]
Merge "Consolidated build inter predictors"

13 years agobug fix removed inline from recon_wrapper_sse2.c
James Berry [Thu, 28 Apr 2011 19:12:00 +0000 (15:12 -0400)]
bug fix removed inline from recon_wrapper_sse2.c

removed inline from recon_wrapper_sse2.c to build
for visual stuido

Change-Id: I74a3482950448e2cdb30e9cd7087145b440d8a22

13 years agobug fix 32 bit matches 64 bit
James Berry [Thu, 28 Apr 2011 18:11:32 +0000 (14:11 -0400)]
bug fix 32 bit matches 64 bit

included vpx_config.h in vpx_encoder.c
to properly define FLOATING_POINT_INIT()

Change-Id: Ie518bf5c087622658e37fca90aa4ddfe79d053f6

13 years agoMerge "Use psadbw to get the sum of bytes in a line."
Scott LaVarnway [Thu, 28 Apr 2011 14:58:20 +0000 (07:58 -0700)]
Merge "Use psadbw to get the sum of bytes in a line."

13 years agoConsolidated build inter predictors
Scott LaVarnway [Thu, 28 Apr 2011 14:53:59 +0000 (10:53 -0400)]
Consolidated build inter predictors

Code cleanup.

Change-Id: Ic8b0167851116c64ddf08e8a3d302fb09ab61146

13 years agoUse psadbw to get the sum of bytes in a line.
Ronald S. Bultje [Wed, 27 Apr 2011 20:49:21 +0000 (13:49 -0700)]
Use psadbw to get the sum of bytes in a line.

Thanks Jason for pointing that out on #vp8. ;-).

Change-Id: I5330a753e752a8704b78a409597472628e0b26a5

13 years agoRemoved unused code in reconinter
Scott LaVarnway [Wed, 27 Apr 2011 19:25:32 +0000 (15:25 -0400)]
Removed unused code in reconinter

The skip flag is never set by the encoder for SPLITMV.

Change-Id: I5ae6457edb3a1193cb5b05a6d61772c13b1dc506

13 years agoMerge "SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}()."
John Koleszar [Wed, 27 Apr 2011 19:02:55 +0000 (12:02 -0700)]
Merge "SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}()."

13 years agoSSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}().
Ronald S. Bultje [Wed, 27 Apr 2011 17:05:10 +0000 (10:05 -0700)]
SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}().

decoding

before
10.425
10.432
10.423
=10.426

after:
10.405
10.416
10.398
=10.406, 0.2% faster

encoding

before
14.252
14.331
14.250
14.223
14.241
14.220
14.221
=14.248

after
14.095
14.090
14.085
14.095
14.064
14.081
14.089
=14.086, 1.1% faster

Change-Id: I483d3d8f0deda8ad434cea76e16028380722aee2

13 years agovpxenc: remove duplicate --fps from vpxenc usage message
Fritz Koenig [Wed, 27 Apr 2011 18:27:59 +0000 (11:27 -0700)]
vpxenc: remove duplicate --fps from vpxenc usage message

Fixes issue #323

Change-Id: I41c297df37afe186a8425ed2e2a95032069dcb9a

13 years agoUse insertion sort instead of quick sort
Yunqing Wang [Wed, 27 Apr 2011 17:40:39 +0000 (13:40 -0400)]
Use insertion sort instead of quick sort

Insertion sort performs better for sorting small arrays. In real-
time encoding (speed=-5), test on test set showed 1.7% performance
gain with 0% PSNR change in average.

Change-Id: Ie02eaa6fed662866a937299194c590d41b25bc3d

13 years agovpxdec: test for frame corruption
John Koleszar [Wed, 27 Apr 2011 16:04:48 +0000 (12:04 -0400)]
vpxdec: test for frame corruption

This change simply exercises the VP8D_GET_FRAME_CORRUPTED control,
outputting a warning message at the end if the bit was set for any
frames. Should never produce any output for good input.

Change-Id: Idaf6ba8f53660f47763cd563fa1485938580a37d

13 years agoMerge "Speed up VP8DX_BOOL_DECODER_FILL"
John Koleszar [Wed, 27 Apr 2011 16:03:45 +0000 (09:03 -0700)]
Merge "Speed up VP8DX_BOOL_DECODER_FILL"

13 years agoMerge "Update VP8DX_BOOL_DECODER_FILL to better detect EOS"
John Koleszar [Wed, 27 Apr 2011 16:03:24 +0000 (09:03 -0700)]
Merge "Update VP8DX_BOOL_DECODER_FILL to better detect EOS"

13 years agoSpeed up VP8DX_BOOL_DECODER_FILL
John Koleszar [Tue, 26 Apr 2011 16:52:17 +0000 (12:52 -0400)]
Speed up VP8DX_BOOL_DECODER_FILL

The end-of-buffer check is hoisted out of the inner loop. Gives
about 0.5% improvement on x86_64.

Change-Id: I8e3ed08af7d33468c5c749af36c2dfa19677f971

13 years agoUpdate VP8DX_BOOL_DECODER_FILL to better detect EOS
John Koleszar [Tue, 26 Apr 2011 16:36:03 +0000 (12:36 -0400)]
Update VP8DX_BOOL_DECODER_FILL to better detect EOS

Allow more reliable detection of truncated bitstreams by being more
precise with the count of "virtual" bits in the value buffer.
Specifically, the VP8_LOTS_OF_BITS value is accumulated into count,
rather than being assigned, which was losing the prior value,
increasing the required tolerance when testing for the error condition.

Change-Id: Ib5172eaa57323b939c439fff8a8ab5fa38da9b69

13 years agoRefactor calc_iframe_target_size
John Koleszar [Tue, 26 Apr 2011 20:45:30 +0000 (16:45 -0400)]
Refactor calc_iframe_target_size

Combine calc_iframe_target_size, previously only used for forced
keyframes, with calc_auto_iframe_target_size, which handled most
keyframes.

Change-Id: I227051361cf46727caa5cd2b155752d2c9789364

13 years agoMove pick_frame_size() to ratectrl.c
John Koleszar [Mon, 25 Apr 2011 19:02:54 +0000 (15:02 -0400)]
Move pick_frame_size() to ratectrl.c

This is a first step in cleaning up the redundancies between
vp8_calc_{auto_,}iframe_target_size. The pick_frame_size() function is
moved to ratectrl.c, and made to be the primary interface. This means
that the various calc_*_target_size functions can be made private.

Change-Id: I66a9a62a5f9c23c818015e03f92f3757bf3bb5c8

13 years agoMerge "Test vector mismatch fix"
Scott LaVarnway [Tue, 26 Apr 2011 17:12:37 +0000 (10:12 -0700)]
Merge "Test vector mismatch fix"

13 years agoTest vector mismatch fix
Scott LaVarnway [Tue, 26 Apr 2011 13:37:19 +0000 (09:37 -0400)]
Test vector mismatch fix

Fixed test vector mismatch that was introduced
in the "Removed dc_diff from MB_MODE_INFO"
(Ie2b9cdf9e0f4e8b932bbd36e0878c05bffd28931)

Change-Id: I98fa509b418e757b5cdc4baa71202f4168dc14ec

13 years agoMerge "remove simpler_lpf"
Johann [Mon, 25 Apr 2011 21:51:07 +0000 (14:51 -0700)]
Merge "remove simpler_lpf"

13 years agoremove simpler_lpf
Johann [Mon, 25 Apr 2011 21:37:41 +0000 (17:37 -0400)]
remove simpler_lpf

the decision to run the regular or simple loopfilter is made outside the
function and managed with pointers

stop tracking the option in two places. use filter_type exclusively

Change-Id: I39d7b5d1352885efc632c0a94aaf56b72cc2fe15

13 years agoFix duplicate vp8_compute_frame_size_bounds
John Koleszar [Mon, 25 Apr 2011 18:30:57 +0000 (14:30 -0400)]
Fix duplicate vp8_compute_frame_size_bounds

Likely introduced by a bad automatic merge from gerrit.

Change-Id: I0c6dd6ec18809cf9492f524d283fa4a3a8f4088b

13 years agoMerge "Remove unused functions"
John Koleszar [Mon, 25 Apr 2011 18:05:00 +0000 (11:05 -0700)]
Merge "Remove unused functions"

13 years agoRemove unused functions
John Koleszar [Mon, 25 Apr 2011 17:54:20 +0000 (13:54 -0400)]
Remove unused functions

Remove estimate_min_frame_size() and calc_low_ss_err(), as they are
never referenced.

Change-Id: I3293363c14ef70b79c4678ca27aa65b345077726

13 years agoMerge "Change rc undershoot/overshoot semantics"
John Koleszar [Mon, 25 Apr 2011 17:49:32 +0000 (10:49 -0700)]
Merge "Change rc undershoot/overshoot semantics"

13 years agoMerge "Stereo 3D format support for vpxenc"
John Koleszar [Mon, 25 Apr 2011 17:48:44 +0000 (10:48 -0700)]
Merge "Stereo 3D format support for vpxenc"

13 years agoMerge "Limit size of initial keyframe in one-pass."
John Koleszar [Mon, 25 Apr 2011 17:48:13 +0000 (10:48 -0700)]
Merge "Limit size of initial keyframe in one-pass."

13 years agoMerge "Add rc_max_intra_bitrate_pct control"
John Koleszar [Mon, 25 Apr 2011 17:47:57 +0000 (10:47 -0700)]
Merge "Add rc_max_intra_bitrate_pct control"

13 years agoLimit size of initial keyframe in one-pass.
John Koleszar [Tue, 19 Apr 2011 18:05:27 +0000 (14:05 -0400)]
Limit size of initial keyframe in one-pass.

Rather than using a default size of 1/2 or 3/2 seconds for the first
frame, use a fraction of the initial buffer level to give the
application some control.

This will likely undergo further refinement as size limits on key
frames are currently under discussion on codec-devel@, but this gives
much better behavior for small buffer sizes as a starting point.

Change-Id: Ieba55b86517b81e51e6f0a9fe27aabba295acab0

13 years agoAdd rc_max_intra_bitrate_pct control
John Koleszar [Mon, 25 Apr 2011 15:44:50 +0000 (11:44 -0400)]
Add rc_max_intra_bitrate_pct control

Adds a control to limit the maximum size of a keyframe, as a function of
the per-frame bitrate. See this thread[1] for more detailed discussion:

[1]: http://groups.google.com/a/webmproject.org/group/codec-devel/browse_thread/thread/271b944a5e47ca38

Change-Id: I7337707642eb8041d1e593efc2edfdf66db02a94

13 years agoMerge "bug fix possible keyframe context divide by zero"
John Koleszar [Mon, 25 Apr 2011 16:35:12 +0000 (09:35 -0700)]
Merge "bug fix possible keyframe context divide by zero"

13 years agobug fix possible keyframe context divide by zero
James Berry [Fri, 22 Apr 2011 15:54:18 +0000 (11:54 -0400)]
bug fix possible keyframe context divide by zero

vp8_adjust_key_frame_context() divides by
estimate_keyframe_frequency() which can
return 0 in the case where --kf-max-dist=0.

Change-Id: Idfc59653478a0073187cd2aa420e98a321103daa

13 years agoStereo 3D format support for vpxenc
Alok Ahuja [Thu, 21 Apr 2011 07:50:07 +0000 (00:50 -0700)]
Stereo 3D format support for vpxenc

Create a new input parameter to allow specifying
the packed frame stereo 3d format. A default value
of mono will be written in the absence of user
specified input

Change-Id: I576d9952ab5d7e2076fbf1b282016a9a1baaa103

13 years agoMerge "keep values in registers during quantization"
Johann [Mon, 25 Apr 2011 13:52:38 +0000 (06:52 -0700)]
Merge "keep values in registers during quantization"

13 years agoMerge "Removed unnecessary frame type checks"
Scott LaVarnway [Mon, 25 Apr 2011 13:45:43 +0000 (06:45 -0700)]
Merge "Removed unnecessary frame type checks"

13 years agoMerge "Removed dc_diff from MB_MODE_INFO"
Scott LaVarnway [Mon, 25 Apr 2011 13:45:32 +0000 (06:45 -0700)]
Merge "Removed dc_diff from MB_MODE_INFO"

13 years agoMerge "make two compiler options explicit for Visual Studio projects"
Yaowu Xu [Fri, 22 Apr 2011 21:08:08 +0000 (14:08 -0700)]
Merge "make two compiler options explicit for Visual Studio projects"

13 years agoFix overflow in temporal_filter_apply_sse2().
Ronald S. Bultje [Thu, 21 Apr 2011 20:35:02 +0000 (16:35 -0400)]
Fix overflow in temporal_filter_apply_sse2().

The accumulator array is an integer array, so use paddd instead of paddw
to add values to it. Fixes overflows when using large --arnr-maxframes
(>8) values.

Change-Id: Iad83794caa02400a65f3ab5760f2517e082d66ae

13 years agoMerge "Remove unused kf rate variables"
John Koleszar [Thu, 21 Apr 2011 23:54:14 +0000 (16:54 -0700)]
Merge "Remove unused kf rate variables"

13 years agoCorrected format specifiers in debug print statements
Adrian Grange [Thu, 21 Apr 2011 22:45:57 +0000 (15:45 -0700)]
Corrected format specifiers in debug print statements

The arguments to these fprintfs are int not long int so
the format specifier should be "%d" and not "%ld". This
was writing garbage in the linux build.

Change-Id: I3d2aa8a448d52e6dc08858d825bf394929b47cf3

13 years agomake two compiler options explicit for Visual Studio projects
Yaowu Xu [Thu, 21 Apr 2011 20:27:42 +0000 (13:27 -0700)]
make two compiler options explicit for Visual Studio projects

This patch changes the release configuration of MS VS projects to
explicitly use two compiler options "Maximize Speed (/O2)" and
"Favor fast code(/Ot)".

Change-Id: I0bf8343d9ca195851332b91ec69c69ee4e31ce2a

13 years agokeep values in registers during quantization
Johann [Wed, 13 Apr 2011 20:38:02 +0000 (16:38 -0400)]
keep values in registers during quantization

add an sse4 quantizer so we can use pinsrw/pextrw and keep values in xmm
registers instead of proxying through the stack. and as long as we're
bumping up, use some ssse3 instructions in the EOB detection (see ssse3
fast quantizer)
pick up about a percent on 32bit and about two on 64bit.

Change-Id: If15abba0e8b037a1d231c0edf33501545c9d9363

13 years agoRemoved unnecessary frame type checks
Scott LaVarnway [Thu, 21 Apr 2011 18:59:42 +0000 (14:59 -0400)]
Removed unnecessary frame type checks

ref_frame is set to INTRA_FRAME for keyframes.  The B_PRED
mode is only used in intra frames.

Change-Id: I9bac8bec7c736300d47994f3cb570329edf11ec0

13 years agoRemoved dc_diff from MB_MODE_INFO
Scott LaVarnway [Thu, 21 Apr 2011 18:38:36 +0000 (14:38 -0400)]
Removed dc_diff from MB_MODE_INFO

The dc_diff flag is used to skip loopfiltering.  Instead
of setting this flag in the decoder/encoder, we now check
for this condition in the loopfilter.

Change-Id: Ie2b9cdf9e0f4e8b932bbd36e0878c05bffd28931

13 years agoRemoved force_no_skip
Scott LaVarnway [Wed, 20 Apr 2011 19:45:12 +0000 (15:45 -0400)]
Removed force_no_skip

force_no_skip is always set to zero.

Change-Id: I89b61c5e0bee34627a9c07c05f3517e1db76af77

13 years agoRemoved redundant checks of the mode_info_context flags
Scott LaVarnway [Tue, 19 Apr 2011 15:42:15 +0000 (11:42 -0400)]
Removed redundant checks of the mode_info_context flags

Code cleanup.  The build inter predictor functions are
redundantly checking the mode_info_context for either
INTRA_FRAME or SPLITMV.

Change-Id: I4d58c3a5192a4c2cec5c24ab1caf608bf13aebfb

13 years agoRemove unused kf rate variables
John Koleszar [Tue, 19 Apr 2011 20:08:45 +0000 (16:08 -0400)]
Remove unused kf rate variables

Remove tot_key_frame_bits and prior_key_frame_size[] as they were
tracked but never used. Remove intra_frame_target, as it was only
used to initialize prior_key_frame_size.

Refactor vp8_adjust_key_frame_context() some to remove unnecessary
calculations.

Change-Id: Icbc2c83d2b90e184be03e6f9679e678f3a4bce8f

13 years agomodify SAVE_XMM for potential 64bit use
Johann [Fri, 15 Apr 2011 14:05:20 +0000 (10:05 -0400)]
modify SAVE_XMM for potential 64bit use

the win64 abi requires saving and restoring xmm6:xmm15. currently
SAVE_XMM and RESTORE XMM only allow for saving xmm6:xmm7. allow
specifying the highest register used and if the stack is unaligned.

Change-Id: Ica5699622ffe3346d3a486f48eef0206c51cf867

13 years agoMerge "Add save/restore xmm registers in x86 assembly code"
Johann [Tue, 19 Apr 2011 13:32:10 +0000 (06:32 -0700)]
Merge "Add save/restore xmm registers in x86 assembly code"

13 years agoAdd save/restore xmm registers in x86 assembly code
Johann [Thu, 7 Apr 2011 17:17:22 +0000 (13:17 -0400)]
Add save/restore xmm registers in x86 assembly code

Went through the code and fixed it. Verified on Windows.

Where possible, remove dependencies on xmm[67]

Current code relies on pushing rbp to the stack to get 16 byte
alignment. This broke when rbp wasn't pushed
(vp8/encoder/x86/sad_sse3.asm). Work around this by using unaligned
memory accesses. Revisit this and the offsets in
vp8/encoder/x86/sad_sse3.asm in another change to SAVE_XMM.

Change-Id: I5f940994d3ebfd977c3d68446cef20fd78b07877

13 years agoMerge "Use sub-pixel search's SSE in mode selection"
Yunqing Wang [Mon, 18 Apr 2011 20:20:04 +0000 (13:20 -0700)]
Merge "Use sub-pixel search's SSE in mode selection"

13 years agoUse sub-pixel search's SSE in mode selection
Yunqing Wang [Mon, 18 Apr 2011 19:48:34 +0000 (15:48 -0400)]
Use sub-pixel search's SSE in mode selection

Passed SSE from sub-pixel search back to pick_inter_mode
function, which is compared with the encode_breakout to
see if we could skip evaluating the remaining modes.

Change-Id: I4a86442834f0d1b880a19e21ea52d17d505f941d

13 years agoMerge "Handle long delay between video frames in multi-thread decoder(issue 312)"
Yunqing Wang [Mon, 18 Apr 2011 17:11:41 +0000 (10:11 -0700)]
Merge "Handle long delay between video frames in multi-thread decoder(issue 312)"

13 years agoMerge "store quant_shift as an unsigned char"
Johann [Mon, 18 Apr 2011 17:03:40 +0000 (10:03 -0700)]
Merge "store quant_shift as an unsigned char"

13 years agoMerge "Add spin-wait pause intrinsic for Windows x64 platform."
Yaowu Xu [Mon, 18 Apr 2011 16:53:26 +0000 (09:53 -0700)]
Merge "Add spin-wait pause intrinsic for Windows x64 platform."

13 years agoMerge "fixed an overflow in ssim calculation"
Yaowu Xu [Mon, 18 Apr 2011 14:44:33 +0000 (07:44 -0700)]
Merge "fixed an overflow in ssim calculation"

13 years agoRemoved unused timers
Scott LaVarnway [Fri, 15 Apr 2011 19:09:07 +0000 (15:09 -0400)]
Removed unused timers

Change-Id: I209803b9dbed2b2f6d02258fd7a3963a6645f4ab

13 years agoMerge "added -fomit-frame-pointer flag for gcc builds"
John Koleszar [Mon, 18 Apr 2011 13:07:57 +0000 (06:07 -0700)]
Merge "added -fomit-frame-pointer flag for gcc builds"

13 years agoHandle long delay between video frames in multi-thread decoder(issue 312)
Yunqing Wang [Fri, 15 Apr 2011 16:57:15 +0000 (12:57 -0400)]
Handle long delay between video frames in multi-thread decoder(issue 312)

This is reported by m...@hesotech.de (see issue 312):
"The decoder causes an access violation
when you decode the first frame, then make a pause of about
60 seconds and then decode further frames. But only if
vpx_codec_dec_cfg_t.threads> 1.

This is caused by a timeout of WaitForSingleObject.
When I change the definition of VPXINFINITE to INFINITE(0xFFFFFFFF),
the problem is solved."

Reproduced the crash and verified the changes on Windows platform.
This brings the behavior inline with the other platforms using sem_wait().

Change-Id: I27b32f90bce05846ef2684b50f7a88f292299da1

13 years agoMerge "remove dead code, add missing RESTORE_XMM"
Johann [Fri, 15 Apr 2011 20:32:54 +0000 (13:32 -0700)]
Merge "remove dead code, add missing RESTORE_XMM"

13 years agoadded -fomit-frame-pointer flag for gcc builds
Scott LaVarnway [Fri, 15 Apr 2011 19:59:21 +0000 (15:59 -0400)]
added -fomit-frame-pointer flag for gcc builds

According to the docs, this should have been enabled, but
the disassembled output shows otherwise.  This improved
the encode/decode performance.

Change-Id: I45ad7e6d299b89ac3166d7ef7da75b74994344c6

13 years agoremove executable bit
Johann [Fri, 15 Apr 2011 17:43:24 +0000 (13:43 -0400)]
remove executable bit

source files are not executable

Change-Id: Id2c7294695a22217468426423979f68f02d82340

13 years agoMerge "Fix usage of value returned by vp8_pick_intra4x4mby_modes"
Adrian Grange [Fri, 15 Apr 2011 15:37:19 +0000 (08:37 -0700)]
Merge "Fix usage of value returned by vp8_pick_intra4x4mby_modes"

13 years agoMerge "Reduce unnecessary distortion computation"
Yunqing Wang [Fri, 15 Apr 2011 15:17:03 +0000 (08:17 -0700)]
Merge "Reduce unnecessary distortion computation"

13 years agoremove dead code, add missing RESTORE_XMM
Johann [Fri, 15 Apr 2011 14:11:53 +0000 (10:11 -0400)]
remove dead code, add missing RESTORE_XMM

vp8_filter_block1d16_h4_ssse3 was never called

because UNSHADOW_ARGS moves the stack by 'mov rsp, rbp', the issue was
masked. however, if/when win64 used those registers for persistant data,
issues could/will arise.

Change-Id: I56d6effca0aeba1f86082689771cb10145d39651

13 years agoFix off-by-one in copy_and_extend_plane
John Koleszar [Fri, 15 Apr 2011 12:44:39 +0000 (08:44 -0400)]
Fix off-by-one in copy_and_extend_plane

Should only copy h lines, not h+1.

Change-Id: I802a85686635900459c6dc79596189033e5298d8

13 years agoReduce unnecessary distortion computation
Yunqing Wang [Thu, 14 Apr 2011 19:53:33 +0000 (15:53 -0400)]
Reduce unnecessary distortion computation

In vp8_pick_inter_mode(), for NEWMV mode, use the error result got
from motion search as distortion. This helps performance in real-
time mode.

Change-Id: I398c4e46cc5381f7d874e748cf78827ef0e0860c

13 years agoMerge "Refactor lookahead ring buffer"
John Koleszar [Thu, 14 Apr 2011 19:35:01 +0000 (12:35 -0700)]
Merge "Refactor lookahead ring buffer"

13 years agoMerge "Use consistent delimiters."
Fritz Koenig [Thu, 14 Apr 2011 18:56:18 +0000 (11:56 -0700)]
Merge "Use consistent delimiters."

13 years agoFix usage of value returned by vp8_pick_intra4x4mby_modes
Adrian Grange [Thu, 14 Apr 2011 17:31:56 +0000 (10:31 -0700)]
Fix usage of value returned by vp8_pick_intra4x4mby_modes

The value of distortion2 returned by vp8_pick_intra4x4mby_modes
was being overwritten by the value returned by get16x16prederror
before it was tested.

Change-Id: If00e80332b272c5545c3a7e381c8041e8319b41a

13 years agoMerge "update configure for ios sdk 4.3"
Johann [Thu, 14 Apr 2011 15:55:22 +0000 (08:55 -0700)]
Merge "update configure for ios sdk 4.3"

13 years agoupdate configure for ios sdk 4.3
Joshua Bleecher Snyder [Thu, 14 Apr 2011 13:28:49 +0000 (09:28 -0400)]
update configure for ios sdk 4.3

update for the latest version of the ios sdk. adding
usr/lib/system fixes a missing libcache.dylib issue

make isysroot path more DRY

Change-Id: Ib748ef3dac3cac2e4848fbffa1e9a0112eac826b

13 years agoUse consistent delimiters.
Fritz Koenig [Wed, 13 Apr 2011 22:06:17 +0000 (15:06 -0700)]
Use consistent delimiters.

opsnr.stt file was using \t for delimiters on everything
except between VPXSSIM and Time.

Change-Id: I6284c4e40c05ff642bf4b0170dca062c279a42df

13 years agoFixed use of early breakout in vp8_pick_intra4x4mby_modes
Adrian Grange [Wed, 13 Apr 2011 19:56:46 +0000 (12:56 -0700)]
Fixed use of early breakout in vp8_pick_intra4x4mby_modes

Index i is used to detect early breakout from the first loop, but
its value is lost due to reuse in the second for loop. I moved
the position of the second loop and did some format cleanup.

Change-Id: I02780eae1bd89df4b6c000fb8a018b0837aac2e5

13 years agoRefactor lookahead ring buffer
John Koleszar [Wed, 13 Apr 2011 18:00:18 +0000 (14:00 -0400)]
Refactor lookahead ring buffer

This patch cleans up the source buffer storage and copy mechanism to
allow access through a standard push/pop/peek interface. This approach
also avoids an extra copy in the case where the source is not a
multiple of 16, fixing issue #102.

Change-Id: I05808c39f5743625cb4c7af54cc841b9b10fdbd9

13 years agostore quant_shift as an unsigned char
Johann [Wed, 13 Apr 2011 17:45:58 +0000 (13:45 -0400)]
store quant_shift as an unsigned char

in encodframe.c, quant_shift is set to 0 or 1 in vp8cx_invert_quant

only use 8 bits to store this, instead of 16. will allow saving an
xmm register in an updated version of the regular quantize

Change-Id: Ie88c47fe2aff5af0283dab1147fb2791e4b12f90

13 years agoChange rc undershoot/overshoot semantics
John Koleszar [Mon, 11 Apr 2011 15:29:23 +0000 (11:29 -0400)]
Change rc undershoot/overshoot semantics

This patch changes the rc_undershoot_pct and rc_overshoot_pct controls
to set the "aggressiveness" of rate adaptation, by limiting the
amount of difference between the target buffer level and the actual
buffer level which is applied to the target frame rate for this frame.

This patch was initially provided by arosenberg at logitech.com as
an attachment to issue #270. It was modified to separate these controls
from the other unrelated modifications in that patch, as well as to
use the pre-existing variables rather than introducing new ones.

Change-Id: Id542e3f5667dd92d857d5eabf29878f2fd730a62