Alex Converse [Mon, 9 May 2016 18:21:20 +0000 (11:21 -0700)]
pickmode: Fix a pair of unsigned overflows.
block_variance: This operates on 8x8s and would be safe with a int32 *
int32 to uint32 multiply, but this is potentially unsafe for 12-bit
input. Unfortunately the code already segfaults on 12-bit input:
https://bugs.chromium.org/p/webm/issues/detail?id=1223
calculate_variance: This operates on up to a 32x32 of 8x8s and can
overflow even with 8-bit input (log2((256*32*32)**2) == 36).
James Zern [Fri, 6 May 2016 02:25:29 +0000 (02:25 +0000)]
Merge changes from topic 'missing-proto'
* changes:
vp9_frame_scale_ssse3.c: make 2 functions static
vp9_pickmode.c: make function static
vp9_noise_estimate.c: make function static
vp9_aq_360.c: add missing include
vp9_idct_intrin_sse2: add missing vp9_rtcd.h include
vpx_dsp/*.[hc]: add missing vpx_dsp_rtcd.h include
Marco [Fri, 29 Apr 2016 22:52:45 +0000 (15:52 -0700)]
vp9: Reduce qp threshold for limiting cyclic refresh on steady blocks.
Makes the delta-qp stop little earlier on areas that have been refreshed enough.
This helps to reduce some pulsing artifact on noisy flat areas observed in some
noisy vc-clips.
Threshold changes only take effect for sources where noise level is estimated to
be >= medium level.
Marco [Thu, 5 May 2016 01:11:19 +0000 (18:11 -0700)]
vp9: Modify logic for gf setting based on up-coming key frame.
For 1 pass vbr mode:
Refactor to move the logic for gf setting based on up-coming
key frames to a separate function, so same logic can be used for
scene-cuts/changes.
Brion Vibber [Tue, 3 May 2016 07:03:59 +0000 (03:03 -0400)]
Skip inttypes.h on Darwin
When building a dynamic framework with Swift compatibility, can't
include any headers that aren't in another module or you get an
error like this from Xcode on the including project:
Include of non-modular header inside framework
For some reason the system inttypes.h is not in a module, unlike
other standard C library headers... but it doesn't seem to be
actually needed on Darwin, so removing it doesn't appear to
be a problem.
Yaowu Xu [Wed, 4 May 2016 19:12:02 +0000 (12:12 -0700)]
Change to use proper type in vp{9,10}_token_state
"qc" in vp{9,10}_token_state is used to save quantized coefficients, this
commit changes the type from short to tran_low_t to properly reflect
the value range for highbitdepth build.
This fixes an out-of-range bug when optimize_b is used in highbitdepth
build.
Jim Bankoski [Tue, 3 May 2016 23:23:06 +0000 (16:23 -0700)]
libvpx: add a unit test for plane_add_noise.
In so doing this fixes a couple of bugs:
vpx_plane_add_noise.c needed to subtract a clamp instead of add.
And the assembly (mmx sse) had assumptions that parameters were
continuous in memory which was not true.
Tom Finegan [Mon, 2 May 2016 17:42:00 +0000 (10:42 -0700)]
configure.sh: Remove armv6-darwin target.
- iOS SDKs no longer ship with armv6 support.
- Our minimum iOS version means all target devices have neon.
- Remove armv6 darwin LD workaround.
- This removes a TODO.
Move skin superblock force split out of this function as well
as some minor code refactors. Checked bitexact for different speed
settings and different resolutions.
First, we only set use_4x4_partition for key frame where we don't
denoise; second, envision we have small partitions, we should pass the
actual block size to denoiser and make an early termination if needed.
Alex Converse [Wed, 27 Apr 2016 04:38:04 +0000 (21:38 -0700)]
Declare the operator== overload for M128A in the global namespace.
ADL will look this up from the callsite namespace iff it is declared
before the callsite or from the parent namespace of the class type (the
global namespace).
This patch has been tested on MSVS 2015 and clang-3.8.
Block size passed into denoiser filter is always >= BLOCK_8X8 (in
vp9_pick_inter_mode), it is not necessary to check smaller block
size. Passed the bitexact test on clips with different resolutions and
noise levels.
Marco [Mon, 25 Apr 2016 21:34:41 +0000 (14:34 -0700)]
vp9: Fix condition to update consec_zero_mv.
Fix will reset the consec_zero_mv map on non-skipped blocks with non-zero mv.
Adjust thresholds on consec_zero_mv in noise estimation and skin detection,
as more possible reset on map means lower thresholds should be used.
webm_read_frame is the only function now which requires
documentation for what the return value means (other two are quite
obvious - file_is_webm and webm_guess_framerate).
vp9: Fix a bug where best_sse_mv is not set properly.
In VP9 internal denoiser, motion magnitude is computed from
best_sse_mv, which should be set to 0 at the begining. This bug may
cause visual aritifact in denoiser. Also, delete two improper comments.
Marco [Mon, 25 Apr 2016 15:49:08 +0000 (08:49 -0700)]
vp9-denoiser: Bugfix and some adjustments for high noise case.
Need to check that sse for non-zero mv has been set for the current block
(i.e., check that nonzero-mv is tested as a mode, so newmv_sse != UINT_MAX)
before forcing to not use zero-mv for denoising.
Also increase some thresholds (sse and sse_diff) for high noise case,
and use shift operaton instead of multiplication on a threshold computation.
paulwilkins [Fri, 11 Mar 2016 10:44:16 +0000 (10:44 +0000)]
Experiment to adapt for net AQ offset.
In Aq mode 1 the segment and AQ delta for each block is based
on spatial variance. There may be a net imbalance between blocks
that have lower Q than the baseline value and those that have higher Q.
This patch monitors that imbalance and extends the allowed baseline
Q range for the frame to accommodate adjustment of that baseline value
to compensate.
projects / libvpx / log