jackychen [Fri, 20 May 2016 20:45:46 +0000 (13:45 -0700)]
vp9: Skip some modes when variance is low for big blocks, for 1 pass real-time.
Skip intra-mode and some inter-modes (newmv, nearmv, nearestmv) for
golden frame if the variance got from choose_partitioning is very low.
Only for 1 pass real-time CBR mode and bsize >= 32x32, it has ~2.5%
speed up with less than 0.1% PSNR drop for rtc test set. Don't see
visual regression.
Linfeng Zhang [Tue, 31 May 2016 17:38:01 +0000 (10:38 -0700)]
Update filter_selectively_vert_row2()
Reduce operations and jumps. perf shows CPU time reduced from 1.9% to
1.6% when decoding fdJc1_IBKJA.248.webm on Xeon E5.
Will apply the changes to vp10 after code review.
Marco [Tue, 31 May 2016 16:37:26 +0000 (09:37 -0700)]
vp9: Skip computation of best_sad for newmv, unless needed.
For non-rd pickmode:
best_pred_sad, computed for NEWMV-last, is only used for
skipping golden non-zero modes. Add condition to avoid this
computation if not used (i.e, if golden nonzero modes are not used).
And remove code for computing best_pred_sad for NEWMV-golden,
since that sad is not used.
No change in behavior; small speed gain (~1%) for svc encodes.
Linfeng Zhang [Tue, 17 May 2016 19:42:55 +0000 (12:42 -0700)]
Upgrade fwht4x4_mmx() to fwht4x4_sse2() for vp9 and vp10.
Function level timing test shows about 27% time saving on
a Xeon E5-2680 v2 desktop.
Rename vp9_dct_sse2.c to vp9_dct_intrin_sse2.c for vp9 and
rename dct_sse2.c to dct_intrin_sse2.c for vp10 to avoid
duplicate basenames.
Actually vp9_fwht4x4_mmx/sse2() and vp10_fwht4x4_mmx/sse2()
are identical. TODO: They should be unified later if there is
no intention to keep a duplicate.
Tom Finegan [Fri, 27 May 2016 16:16:35 +0000 (09:16 -0700)]
vpx_ports/mem_ops.h: cast the lhs of bitwise shifts of 24.
C does not allow for shifting into the sign bit of a signed
integer, and the two instances here become signed ints via
promotion. Explcitly cast them to unsigned MEM_VALUE_T to
avoid the problem.
Linfeng Zhang [Thu, 19 May 2016 23:52:14 +0000 (16:52 -0700)]
Upgrade vpx_lpf_{vertical,horizontal}_4 mmx to sse2
Followed the code style of other lpf fuctions.
These 2 functions put 2 rows of data in a single xmm register,
so they have similar but not identical filter operations,
and cannot share the same macros.
Brion Vibber [Tue, 24 May 2016 18:42:51 +0000 (11:42 -0700)]
Move git version extras out of iOS shared framework bundle version
Apple's version format specification is strictly checked on app
store submission, even for embedded frameworks:
http://apple.co/1WgelY1
The build version number should be a string comprised of
three non-negative, period-separated integers with the
first integer being greater than zero. The string should
only contain numeric (0-9) and period (.) characters.
The full version returned from 'version.sh --bare' is now
embedded under a 'VPXFullVersion' custom key in the Info.plist,
so it can still be extracted from the resulting framework.
James Zern [Tue, 24 May 2016 19:02:22 +0000 (12:02 -0700)]
remove vp9_diamond_search_sad_avx.c
vp9_diamond_search_sad_avx was disabled in: 057c1c4 disable vp9_diamond_search_sad_avx
this removes a missing prototype warning as the prototype is no longer
included in vp9_rtcd.h. the file can be restored if someone gets around
to fixing the issue.
JackyChen [Mon, 16 May 2016 20:41:56 +0000 (13:41 -0700)]
vp9: Refactor some denoiser logic in vp9_pick_inter_mode.
Move the logic for rechecking zeromv on denoised block out to simplify
the function. To simplify the param passing, add a new structure
VP9_PICKMODE_CTX_DEN which is only used when denoiser is enabled.
Yaowu Xu [Wed, 18 May 2016 17:05:38 +0000 (10:05 -0700)]
Clarify integer value ranges
This commit clarifies integer value range for vairables used in
several variance functions, also change to use proper type
conversion to reflect the value ranges.
Alex Converse [Wed, 18 May 2016 00:45:33 +0000 (17:45 -0700)]
Move, rename, and inline high_inter_predictor.
The inlining mirrors what was done with the low bit depth
inter_predictor. And the new highbd_inter_predictor name is more
consistent with other high bit depth functions.
Brion Vibber [Mon, 2 May 2016 16:41:59 +0000 (12:41 -0400)]
Add --enable-shared option to iosbuild.sh to build dynamic framework
Also allows use of --enable-shared when configuring for Mac OS X,
producing a bare .dylib.
Enabling the shared framework bumps the iOS deployment target to 8.0,
the minimum required to support dynamic framework deployment in apps.
When not using --enable-shared, a static library for iOS 6.0+ will still
be built.
Minimum version settings have been moved into ios-version.sh so they
can be updated in a single place.
As with the static build, unless header search paths are manually
tweaked, users must add a VPX prefix on includes, such as:
#include <VPX/vpx/vpx_decoder.h>
A module map for headers is not yet included as inttypes.h is not
modular; this means that VPX cannot be used directly in Swift code,
but can still be pulled in through an Objective-C wrapper.
Issue introduced in https://chromium-review.googlesource.com/#/c/339162/.
The factor on avg_frame_qindex[INTER_FRAME] in that CL, used for q-basis to
set active_best, can cause the QP to decrease too slowly or get stuck
in some cases when max-q=63.
Removing that factor fixes the regression in issue#1217.
Yaowu Xu [Wed, 11 May 2016 20:52:59 +0000 (13:52 -0700)]
Change to use correct check for halfpel
In motion estimation stage for subpel motion, subpel variance is
computed use bilinear interpolation. The motion vector precision
used is at 1/8 pel and three bits are used to represent the x and y
subpel offsets. Based on this, the half pel check should be against
4, not 8.
Linfeng Zhang [Wed, 11 May 2016 19:39:42 +0000 (12:39 -0700)]
remove mmx variance functions
there are sse2 equivalents which is a reasonable modern baseline
Removed mmx variance functions:
vpx_get_mb_ss_mmx()
vpx_get8x8var_mmx()
vpx_get4x4var_mmx()
vpx_variance4x4_mmx()
vpx_variance8x8_mmx()
vpx_mse16x16_mmx()
vpx_variance16x16_mmx()
vpx_variance16x8_mmx()
vpx_variance8x16_mmx()
paulwilkins [Thu, 5 May 2016 10:37:04 +0000 (11:37 +0100)]
Fixed 8K two pass encoder crash.
Bug found by Yunqing relating to the correction for size at 8K and
above in get_twopass_worst_quality().
The basis for the correction was changed to the linear size relative to
1080P as a baseline and the adjustment has been clamped to prevent
problems at extreme images sizes.
For 1080P the results on our test sets were neutral but the low res and
mid res sets saw a small gain (0.1%-0.2% average).
I would also expect some gains on 4k and larger content where the
previous correction was overly aggressive.
hui su [Thu, 5 May 2016 22:37:37 +0000 (15:37 -0700)]
Add VP9 encoder API for level specification.
Add control API VP9E_SET_TARGET_LEVEL that allows the encoder to
control the output bitstream level and/or keep level related
statistics.
Usage:
255 do not care about level (default)
0 keep level related stats only
10 target for level 1
11 target for level 1.1
.
.
.
62 target for level 6.2
projects / libvpx / log