coffee/libyuv
1
0
Fork 0
mirror of https://chromium.googlesource.com/libyuv/libyuv synced 2025-12-07 01:06:46 +08:00
Code Issues Packages Projects Releases Wiki Activity
1,589 Commits 3 Branches 0 Tags BSD-3-Clause 60 MiB
Commit Graph

256 Commits

Author SHA1 Message Date
Frank Barchard
f4447745ae Add rounding to InterpolateRow for improved quality and consistency. 
Remove inaccurate specializations for 1/4 and 3/4, since they round
incorrectly.  Specialize for 100% and 50% are kept due to performance.
Make C and ARM code match SSSE3.
Make unittests expect zero difference.

BUG=libyuv:535
R=harryjin@google.com

Review URL: https://codereview.chromium.org/1533643005 .
 2015-12-17 15:24:06 -08:00
Frank Barchard
ae55e41851 use rounding in scaledown by 2 
When scaling down by 2 the formula should round consistently.
(a+b+c+d+2)/4
The C version did but the SSE2 version was doing 2 averages.
avg(avg(a,b),avg(c,d))
This change uses a sum, then rounds.

R=dhrosa@google.com, harryjin@google.com
BUG=libyuv:447,libyuv:527

Review URL: https://codereview.chromium.org/1513183004 .
 2015-12-14 17:25:36 -08:00
Frank Barchard
a2ea905679 BlendPlane any width. 
Benchmark
out\release\libyuv_unittest --libyuv_width=1279 --libyuv_height=719 --libyuv_repeat=999 --libyuv_flags=-1 --gtest_filter=*Blend* | sortms

Was
I420Blend_Any (2321 ms)
I420Blend_Unaligned (1684 ms)
I420Blend_Opt (1675 ms)
I420Blend_Invert (1653 ms)
BlendPlane_Invert (1556 ms)
BlendPlane_Any (1552 ms)
BlendPlane_Unaligned (1548 ms)
BlendPlane_Opt (1535 ms)
ARGBBlend_Unaligned (659 ms)
ARGBBlend_Any (596 ms)
ARGBBlend_Invert (591 ms)
ARGBBlend_Opt (508 ms)
BlendPlaneRow_Unaligned (186 ms)
BlendPlaneRow_Opt (171 ms)

Now
ARGBBlend_Any (621 ms)
ARGBBlend_Unaligned (585 ms)
ARGBBlend_Invert (564 ms)
ARGBBlend_Opt (512 ms)
I420Blend_Unaligned (347 ms)
I420Blend_Invert (345 ms)
I420Blend_Any (337 ms)
I420Blend_Opt (327 ms)
BlendPlane_Unaligned (187 ms)
BlendPlaneRow_Unaligned (187 ms)
BlendPlane_Invert (186 ms)
BlendPlane_Any (186 ms)
BlendPlaneRow_Opt (173 ms)
BlendPlane_Opt (171 ms)

which is comparable to aligned case
out\release\libyuv_unittest --libyuv_width=1280 --libyuv_height=720 --libyuv_repeat=999 --libyuv_flags=-1 --gtest_filter=*Blend* | sortms
ARGBBlend_Any (625 ms)
ARGBBlend_Unaligned (602 ms)
ARGBBlend_Invert (508 ms)
ARGBBlend_Opt (506 ms)
I420Blend_Any (353 ms)
I420Blend_Unaligned (322 ms)
I420Blend_Invert (304 ms)
I420Blend_Opt (301 ms)
BlendPlaneRow_Unaligned (188 ms)
BlendPlane_Unaligned (186 ms)
BlendPlane_Invert (185 ms)
BlendPlane_Any (184 ms)
BlendPlaneRow_Opt (173 ms)
BlendPlane_Opt (169 ms)

R=dhrosa@google.com, harryjin@google.com
BUG=libyuv:527

Review URL: https://codereview.chromium.org/1513443002 .
 2015-12-08 18:59:48 -08:00
Frank Barchard
dee77a4ebe Optimize yuv alpha blend AVX2 code to do 32 pixels at time. 
out/Release/libyuv_unittest --libyuv_width=1280 --libyuv_height=720 --libyuv_repeat=9999 --libyuv_flags=-1 --gtest_filter=*I420Blend_Opt

Was LibYUVPlanarTest.I420Blend_Opt (2335 ms)
Now LibYUVPlanarTest.I420Blend_Opt (1937 ms)

vs SSSE3
LibYUVPlanarTest.I420Blend_Opt (2599 ms)

BUG=libyuv:527
R=dhrosa@google.com

Review URL: https://codereview.chromium.org/1505673003 .
 2015-12-08 18:20:30 -08:00
Frank Barchard
2657688e70 Add support for odd height YUVA alpha blending. 
R=dhrosa@google.com, harryjin@google.com
BUG=libyuv:527

Review URL: https://codereview.chromium.org/1507683003 .
 2015-12-07 12:03:20 -08:00
Frank Barchard
48a919d86e Bug fix for UYVYToNV12 odd height 
TBR=harryjin@google.com
BUG=libyuv:528

Review URL: https://codereview.chromium.org/1506973002 .
 2015-12-07 11:39:48 -08:00
Frank Barchard
bea690b3e0 AVX2 YUV alpha blender and improved unittests 
AVX2 version can process 16 pixels at a time for improved memory bandwidth and fewer instructions.

unittests improved to test unaligned memory, and test exactness when alpha is 0 or 255.

R=dhrosa@google.com, harryjin@google.com
BUG=libyuv:527

Review URL: https://codereview.chromium.org/1505433002 .
 2015-12-05 22:23:29 -08:00
Frank Barchard
b6f37bd8ec Interpolate plane initial implementation. 
YUV version of interpolation between two images.

R=dhrosa@google.com, harryjin@google.com
BUG=libyuv:526

Review URL: https://codereview.chromium.org/1479593002 .
 2015-11-25 16:11:42 -08:00
Frank Barchard
d95d2169d9 rename yuv matrix constants to be more clear about what they are 
R=harryjin@google.com
BUG=none

Review URL: https://codereview.chromium.org/1429693006 .
 2015-11-03 17:09:53 -08:00
Frank Barchard
ce4c2fad1d Raw 24 bit RGB to RGB24 (bgr) 
Add unittests that do 1 step conversion vs 2 step conversion.

Tests end swapping versions match direct conversions.

R=harryjin@google.com
BUG=libyuv:518

Review URL: https://codereview.chromium.org/1419103007 .
 2015-11-03 10:30:30 -08:00
Frank Barchard
2c7aa0070a remove I422ToBGRA and use I422ToRGBA internally 
Removes low levels for I420ToBGRA and I420ToRAW and reimplements them as I420ToRGBA and I420ToRGB24 with transposed color matrix.

Adds unittests that do 1 step conversion vs 2 steps to test end swapping versions match direct conversions.

R=harryjin@google.com
BUG=libyuv:518

Review URL: https://codereview.chromium.org/1427993004 .
 2015-11-02 10:24:12 -08:00
Frank Barchard
5d97b93369 refactor I420ToABGR to use I420ToARGBRow 
Using a transposed conversion matrix, I420ToARGB can output ABGR.

R=harryjin@google.com, xhwang@chromium.org
BUG=libyuv:473

Review URL: https://codereview.chromium.org/1413573010 .
 2015-10-30 11:56:57 -07:00
Frank Barchard
76a599ec3b fix jpeg and bt.709 yuvconstants for neon64. 
yuv constants for bt.601 were previously ported to neon64, as well
as the code to respect other color spaces.  But the jpeg and bt.709
colour conversion constants were still in armv7 form.  This changes
the constants for aarch64 builds to be compatible with the code.

yuv constants are now passed as const *

Remove Yvu constants which were used for older version on nv21 but not new code.

TBR=harryjin@google.com
BUG=none

Review URL: https://codereview.chromium.org/1398623002 .
 2015-10-07 19:46:56 -07:00
Frank Barchard
914a9856c7 Reimplement NV21ToARGB to allow different color matrix. 
Low level for NV21ToARGB written to accept yuv matrix used by
other YUV to ARGB functions.
Previously NV21 was implemented for Windows using NV12 with a different
matrix that swapped U and V.  But the Arm version of the low level does
not allow the matrix U and V contributions to be swapped.
Using a new low level function that reads NV21 and uses the same
yuvconstants as other YUV conversion functions allows an Arm port of
this function.

TBR=harryjin@google.com
BUG=libyuv:500

Review URL: https://codereview.chromium.org/1388273002 .
 2015-10-06 20:34:44 -07:00
Frank Barchard
2cc1a2b233 Remove sse2 functions that also have ssse3 
ARGBBlendRow_SSE2, ARGBAttenuateRow_SSE2, and MirrorRow_SSE2
Since vast majority of CPUs have SSSE3 now, removing the SSE2
improves the performance of CPU dispatching.

R=harryjin@google.com
BUG=none

Review URL: https://codereview.chromium.org/1377053003 .
 2015-09-30 14:24:44 -07:00
Frank Barchard
f96890a0be yuvconstants for all YUV to RGB conversion functions. 
R=harryjin@google.com
BUG=libyuv:488

Review URL: https://codereview.chromium.org/1363503002 .
 2015-09-22 10:26:03 -07:00
Frank Barchard
278d88f872 Copy Alpha odd width support 
R=harryjin@google.com
BUG=none

Review URL: https://webrtc-codereview.appspot.com/59369004.
 2015-08-13 15:05:14 -07:00
Frank Barchard
ce98129951 yuy2tonv12 
R=bcornell@google.com
BUG=libyuv:466

Review URL: https://webrtc-codereview.appspot.com/51309004.
 2015-07-17 16:22:59 -07:00
Frank Barchard
faa4b14f85 uyvy to nv12 
R=harryjin@google.com
BUG=libyuv:466

Review URL: https://webrtc-codereview.appspot.com/50339004.
 2015-07-17 14:43:19 -07:00
fbarchard@google.com
bd2d903e1b odd width support for ARGBSobel functions. Improves performance for images that are not a multiple of 8 pixels. 
BUG=444
TESTED=libyuvTest.ARGBSobel_Opt
R=harryjin@google.com

Review URL: https://webrtc-codereview.appspot.com/54589004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1415 16f28f9a-4ce2-e073-06de-1de4eb20be90
 2015-05-28 22:22:28 +00:00
fbarchard@google.com
cfce47efc8 Change Sobel to use JPeg Luma calculation instead of extracting G channel. Using luma produces a better sobel that respects all 3 channels of RGB. Historically the G channel was used to improve performance, and because the luma of I420 is a constrained range, hurting quality. Using the JPeg variation of YUV, the luma is more accurate, including cross platform, better optimized for AVX2 and odd widths, and full range. 
BUG=444
TESTED=ARGBSobelXY_Opt
R=harryjin@google.com

Review URL: https://webrtc-codereview.appspot.com/57479004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1414 16f28f9a-4ce2-e073-06de-1de4eb20be90
 2015-05-27 22:32:26 +00:00
fbarchard@google.com
8f0b32773c ARGBToUV AVX2 functions hooked up. 
BUG=none
TESTED=RGB565ToI420
R=tpsiaki@google.com

Review URL: https://webrtc-codereview.appspot.com/46829004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1359 16f28f9a-4ce2-e073-06de-1de4eb20be90
 2015-04-07 00:10:52 +00:00
fbarchard@google.com
e6ca9cc2a2 Scale down by 2 AVX2 port. Processes twice as many pixels as SSE2 and takes advantage of 3 argument instructions to reduce register usage and number of instructions. 
BUG=314
TESTED=libyuvTest.ScaleDownBy2_Box
R=tpsiaki@google.com

Review URL: https://webrtc-codereview.appspot.com/42959004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1347 16f28f9a-4ce2-e073-06de-1de4eb20be90
 2015-03-26 23:21:08 +00:00
fbarchard@google.com
d28cd77f99 Enable assembly for clangcl build on Windows. Previously assembly was disabled so clangcl would work, but only with C code. As clangcl mimics both Visual C and GCC, ifdefs need to pick one or the other or often you'll end up with both. In this CL we disable most Visual C code and use the GCC versions which allow assembly for both 32 and 64 bit intel. 
BUG=412
TESTED=clang=1 build on windows
R=tpsiaki@google.com

Review URL: https://webrtc-codereview.appspot.com/51389004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1341 16f28f9a-4ce2-e073-06de-1de4eb20be90
 2015-03-19 20:36:31 +00:00
fbarchard@google.com
446fa95587 I422ToRGB565, ARGB4444 and ARGB1555 for AVX2 
BUG=403
TESTED=avx2 emulator

Review URL: https://webrtc-codereview.appspot.com/34359004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1293 16f28f9a-4ce2-e073-06de-1de4eb20be90
 2015-02-24 23:14:46 +00:00
fbarchard@google.com
fd8054791c build fixe for InterpolateRow_MIPS_DSPR2 
BUG=398
TESTED=untested
R=harryjin@google.com

Review URL: https://webrtc-codereview.appspot.com/37999004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1268 16f28f9a-4ce2-e073-06de-1de4eb20be90
 2015-02-07 01:01:30 +00:00
fbarchard@google.com
b2a6af1be6 Change rectangle low level functions to use more conventional row functions including 'any' variations. Previously the yuv function SetPlane stored 32 bit values. Now a more conventional memset() style function is used for YUV that stores bytes. On Haswell a rep stosb is used for YUV. Overall benefit of this CL is improved performance for 'any' width, and simpler row assembly instead of full image assembly. Previously ARGBRect used a low level function that supported a rectangle in assembly. Now it uses a row function, and relies on row coalesce to combine into a single low level call. 
BUG=371
TESTED=untested
R=brucedawson@google.com, harryjin@google.com

Review URL: https://webrtc-codereview.appspot.com/35689004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1222 16f28f9a-4ce2-e073-06de-1de4eb20be90
 2015-01-12 03:58:24 +00:00
fbarchard@google.com
8e3db2dc73 Support invert for ARGBRect and SetPlane 
BUG=387
TESTED=ARGBRect_Invert
R=harryjin@google.com

Review URL: https://webrtc-codereview.appspot.com/37539004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1219 16f28f9a-4ce2-e073-06de-1de4eb20be90
 2015-01-07 19:02:01 +00:00
fbarchard@google.com
992c3b089a Use HAS_ARGBSETROWS_X86 to detect presence of function. 
BUG=none
TESTED=rectangle unittests
R=harryjin@google.com

Review URL: https://webrtc-codereview.appspot.com/35639004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1218 16f28f9a-4ce2-e073-06de-1de4eb20be90
 2015-01-07 00:11:51 +00:00
fbarchard@google.com
ef67597b48 ARGBMirror use SSE2 pshufd instruction instead of SSSE3 pshufb. 
BUG=269
TESTED=local benchmark for ARGBMirror
R=tpsiaki@google.com

Review URL: https://webrtc-codereview.appspot.com/32509004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1176 16f28f9a-4ce2-e073-06de-1de4eb20be90
 2014-11-21 19:25:14 +00:00
fbarchard@google.com
91f240c5db Move sub before branch for loops. 
Remove CopyRow_x86
Add CopyRow_Any versions for AVX, SSE2 and Neon.
BUG=269
TESTED=local build
R=harryjin@google.com, tpsiaki@google.com

Review URL: https://webrtc-codereview.appspot.com/26209004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1175 16f28f9a-4ce2-e073-06de-1de4eb20be90
 2014-11-20 21:14:27 +00:00
fbarchard@google.com
9dd083a512 ARGBMirror Any 
BUG=none
TESTED=mirror and rotate unittests
R=harryjin@google.com

Review URL: https://webrtc-codereview.appspot.com/30159004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1172 16f28f9a-4ce2-e073-06de-1de4eb20be90
 2014-11-19 00:46:51 +00:00
fbarchard@google.com
59ed448685 MirrorAny functions so assembly can always be used. 
BUG=none
TESTED=untested
R=harryjin@google.com

Review URL: https://webrtc-codereview.appspot.com/29069004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1170 16f28f9a-4ce2-e073-06de-1de4eb20be90
 2014-11-18 01:03:47 +00:00
fbarchard@google.com
9ed836b154 The 'Any' versions of functions can handle any width now, so remove the check from the calling code. This has 2 advantages - less code, and less overhead in calling function when any function is NOT used. Downside is more code for case where any is used. 
BUG=373
TESTED=libyuv_unittest still passes
R=harryjin@google.com

Review URL: https://webrtc-codereview.appspot.com/24129004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1143 16f28f9a-4ce2-e073-06de-1de4eb20be90
 2014-10-24 23:29:31 +00:00
fbarchard@google.com
0a6dab42c0 Add check for minimum of 8 pixels for any functions and multiple of 8 not 16 for neon functions. 
BUG=373
TESTED=try bots
R=harryjin@google.com

Review URL: https://webrtc-codereview.appspot.com/23189004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1139 16f28f9a-4ce2-e073-06de-1de4eb20be90
 2014-10-23 23:05:12 +00:00
fbarchard@google.com
f2fa453b94 Port I422ToABGR to AVX2. 
BUG=269
TESTED=intelsde on I422ToABGR
R=tpsiaki@google.com

Review URL: https://webrtc-codereview.appspot.com/23149004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1138 16f28f9a-4ce2-e073-06de-1de4eb20be90
 2014-10-23 17:20:22 +00:00
fbarchard@google.com
c000955bc0 Port I422ToRGBA to AVX. 
BUG=269
TESTED=intelsde on I422ToRGBA
R=brucedawson@google.com

Review URL: https://webrtc-codereview.appspot.com/28769004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1136 16f28f9a-4ce2-e073-06de-1de4eb20be90
 2014-10-22 22:41:39 +00:00
fbarchard@google.com
d81dddd3d0 port I420ToBGRA to AVX2. 
BUG=269
TESTED=c:\intelsde\sde -ast -hsw -- out\release\libyuv_unittest.exe --gtest_filter=*I420ToBGRA*
R=brucedawson@google.com, harryjin@google.com, magjed@chromium.org

Review URL: https://webrtc-codereview.appspot.com/26869004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1127 16f28f9a-4ce2-e073-06de-1de4eb20be90
 2014-10-20 19:35:55 +00:00
fbarchard@google.com
055725b8fe Neon does 8 at a time, so a check is added for any function of I422ToBGRA that width is >= 8 and for fast path that it is a multiple of 8 not 16. 
BUG=373
TESTED=untested
R=brucedawson@google.com

Review URL: https://webrtc-codereview.appspot.com/24059004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1126 16f28f9a-4ce2-e073-06de-1de4eb20be90
 2014-10-20 18:39:21 +00:00
fbarchard@google.com
f713691a6f Change elif to endif and if to allow AVX2 as well as SSE2 in future changes instead of one or the other. 
BUG=none
TESTED=try bots
R=harryjin@google.com

Review URL: https://webrtc-codereview.appspot.com/30719004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1122 16f28f9a-4ce2-e073-06de-1de4eb20be90
 2014-10-16 20:47:22 +00:00
fbarchard@google.com
b720049a54 Make row functions used for planarfunctions and convert use movdqu to relax alignment constraint. Step 1 - make functions unaligned. 
BUG=365
TESTED=libyuv_unittest passes
R=harryjin@google.com

Review URL: https://webrtc-codereview.appspot.com/26709004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1111 16f28f9a-4ce2-e073-06de-1de4eb20be90
 2014-10-03 21:11:37 +00:00
fbarchard@google.com
044f914c29 Change scale to unaligned movdqu. 
BUG=365
TESTED=scale unittests
R=tpsiaki@google.com

Review URL: https://webrtc-codereview.appspot.com/22879004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1101 16f28f9a-4ce2-e073-06de-1de4eb20be90
 2014-10-01 01:16:04 +00:00
fbarchard@google.com
d33bf86b25 CopyRow_AVX which supports unaligned pointers for Sandy Bridge CPU. 
BUG=363
TESTED=out\release\libyuv_unittest --gtest_filter=*ARGBToARGB_*
R=tpsiaki@google.com

Review URL: https://webrtc-codereview.appspot.com/31489004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1097 16f28f9a-4ce2-e073-06de-1de4eb20be90
 2014-09-29 23:53:18 +00:00
fbarchard@google.com
a9ff15b7bb check copy has different address. If same, skip the copy to avoid valgrind error. 
BUG=334
TESTED=unittests still pass
R=tpsiaki@google.com

Review URL: https://webrtc-codereview.appspot.com/14679004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1011 16f28f9a-4ce2-e073-06de-1de4eb20be90
 2014-06-11 00:16:59 +00:00
fbarchard@google.com
b18413e568 YUV scaling with 16 bit planes 
BUG=331
TESTED=libyuv_unittest --gunit_also_run_disabled_tests --gtest_filter=**.ScaleFrom1280x720*
R=debargha@google.com, tpsiaki@google.com

Review URL: https://webrtc-codereview.appspot.com/17569004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1005 16f28f9a-4ce2-e073-06de-1de4eb20be90
 2014-05-20 19:22:30 +00:00
fbarchard@google.com
70bc4995a0 Planarfunctions (mainly effects) converted to C89/VisualC. 
BUG=303
TESTED=cl /c /TC /Iinclude source/planar_functions.cc
R=tpsiaki@google.com

Review URL: https://webrtc-codereview.appspot.com/6979004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@965 16f28f9a-4ce2-e073-06de-1de4eb20be90
 2014-01-13 18:56:25 +00:00
fbarchard@google.com
1f923e3ea6 Declare parameters that are unused, since C does not let you give a type without name. 
BUG=303
TEST=compile -x c

Review URL: https://webrtc-codereview.appspot.com/6599006

git-svn-id: http://libyuv.googlecode.com/svn/trunk@954 16f28f9a-4ce2-e073-06de-1de4eb20be90
 2014-01-07 05:42:27 +00:00
fbarchard@google.com
a1f5254a95 Switch to c style casts for all source and includes. 
BUG=303
TESTED=try
R=tpsiaki@google.com

Review URL: https://webrtc-codereview.appspot.com/6629004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@952 16f28f9a-4ce2-e073-06de-1de4eb20be90
 2014-01-07 03:03:00 +00:00
fbarchard@google.com
a12284b906 sobel use one alloc instead of 3. 
BUG=300
TESTED=try bots
R=tpsiaki@google.com

Review URL: https://webrtc-codereview.appspot.com/6449004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@931 16f28f9a-4ce2-e073-06de-1de4eb20be90
 2013-12-30 18:39:43 +00:00
fbarchard@google.com
49db7b7e4a Add edge to sobel buffers to avoid overwrites. 
BUG=296
TESTED=Sobel unittest in Effects
R=tpsiaki@google.com

Review URL: https://webrtc-codereview.appspot.com/6429004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@930 16f28f9a-4ce2-e073-06de-1de4eb20be90
 2013-12-27 22:54:51 +00:00