Inline that uses temporary variables is currently initializing them
to 0 and passing in as output "+r".
This CL replaces the output constraint to "=&r" for most meaning an
output with early write (before inputs). This allows the initialize
to zero step to be removed, saving 1 instruction.
BUG=libyuv:580
TESTED=local libyuv build on gcc/linux and try bots
R=harryjin@google.com
Review URL: https://codereview.chromium.org/1895743008 .
When using C version of I420Interpolate for msan, a 50% interpolation
would cause stride to be cast to int, which could cause erroneous
memory reads on 64 bit build.
This CL makes the stride use ptrdiff_t for HalfRow_C
BUG=libyuv:582
TESTED=try bots tests
R=dhrosa@google.com
Review URL: https://codereview.chromium.org/1872953002 .
An SSSE3 version already exists, and an AVX2 version is available for
Visual C. This ports the function to AVX2 completing the AVX2 ports of
all YUV to RGB functions for AVX2 on gcc.
TBR=harryjin@google.com
BUG=libyuv:555
Review URL: https://codereview.chromium.org/1687253002 .
When attempting to normalize function names to end in Row_SIMD it was made
harder with MIPS_DSPR2 naming convention.
Other CPUs do not include the vendor. This should be named consistently.
Removed the DISABLE_MIPS in favour of DISABLE_ASM for consistency with other
processors.
TBR=harryjin@google.com
BUG=libyuv:562
Review URL: https://codereview.chromium.org/1677633002 .
internal math of the fastrand function uses a multiply
and add that overflows a signed int. This triggers a
ubsan failure:
../../unit_test/../unit_test/unit_test.h:60:33: runtime error: signed integer overflow: 56248274 * 214013 cannot be represented in type 'int'
This change casts the intermediate math to unsigned
int to avoid the overflow.
For more info on ubsan, see
http://dev.chromium.org/developers/testing/undefinedbehaviorsanitizer
TESTED=Passing compilation using:
GYP_DEFINES="ubsan=1"
GYP_DEFINES="ubsan_vptr=1"
R=harryjin@google.com, pbos@webrtc.org
BUG=libyuv:563
Review URL: https://codereview.chromium.org/1662453003 .
This is an UBSan error reported by libjingle
[ RUN ] WebRtcVideoFrameTest.ConvertToYUY2BufferStride
[000:000] (videoframe.cc:375): Validate frame passed. format: I420 bpp: 12 size: 1280x720 bytes: 1382400 expected: 1382400 sample[0..3]: 73, 73, 73, 73
../../chromium/src/third_party/libyuv/source/row_gcc.cc:2903:25: runtime error: signed integer overflow: 128 * 16843009 cannot be represented in type 'int'
[8/614] WebRtcVideoFrameTest.ConvertToYUY2BufferStride returned/aborted with exit code 1 (32 ms)
[9/614] WebRtcVideoFrameTest.ConvertToYUY2BufferInverted (29 ms)
Note: Google Test filter = WebRtcVideoFrameTest.ConvertToYUY2BufferInverted
The source is uint8 and the multiply is by 0x01010101 to replicate the byte to 4 bytes.
Changing the constant to 0x01010101u should avoid overflow.
R=harryjin@google.comTBR=harryjin@google.com
BUG=libyuv:563
Review URL: https://codereview.chromium.org/1657533005 .
When the image height for unittests was set to an
odd height, the TestI420 unittest would not fill
the complete source buffer. This change handles
the odd height test case.
No change to library code.
TBR=harryjin@google.com
BUG=libyuv:549
Review URL: https://codereview.chromium.org/1609103002 .
When width was odd Y channel wrote an extra pixel.
This change splits the Y from UV into a temporary
buffer and memcpy's to the destination. Performance
is slower.
Was
YUY2ToNV12_Any (307 ms)
YUY2ToNV12_Unaligned (213 ms)
TestYUY2ToNV12 (181 ms)
YUY2ToNV12_Opt (177 ms)
YUY2ToNV12_Invert (177 ms)
Npw
YUY2ToNV12_Any (300 ms)
YUY2ToNV12_Unaligned (226 ms)
YUY2ToNV12_Invert (206 ms)
TestYUY2ToNV12 (184 ms)
YUY2ToNV12_Opt (181 ms)
TBR=harryjin@google.com
BUG=libyuv:545
Review URL: https://codereview.chromium.org/1593833002 .
I420ToNV21 passes the wrong dst_stride_y when it calls I420ToNV12; parameter 8 (convert_from.cc:448) is src_stride_y but should be dst_stride_y. This causes image corruption when converting I420 -> NV21 with mismatched luminance strides.
R=dhrosa@google.com, harryjin@google.com
BUG=libyuv:547
Review URL: https://codereview.chromium.org/1582793008 .
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 .
Some functions run out of registers when compiled for debug,
fpic, with stack frames on 32 bit x86 with clang.
Previously they were enabled based on _DEBUG but that macro
is not set in some build systems. This CL adds DEBUG macro as
well to cover those environments.
R=harryjin@google.com
BUG=libyuv:532
Review URL: https://codereview.chromium.org/1517693005 .
previously the I411 format used movd to read U, V pixels.
But this reads 4 bytes, and can cause a memory exception.
pinsrw can be used, but fails on drmemory 1.5, and is slow.
So in this change a movzxw is used to read 2 bytes into EBX,
then copy to xmm0 with movd.
Slightly slower, but no memory exception
Was LibYUVConvertTest.I411ToARGB_Opt (577 ms)
Now LibYUVConvertTest.I411ToARGB_Opt (608 ms)
TBR=harryjin@google.com
BUG=libyuv:525
Review URL: https://codereview.chromium.org/1457783004 .
libyuv builds/runs, but when integrated into chromium, produces link errors. unclear why but this disables affected functions.
will followup with re-enabling them once the root cause in the runtime error is found.
TBR=harryjin@google.com
BUG=libyuv:522
Review URL: https://codereview.chromium.org/1427683004 .
disable I422ALPHATOARGBROW_AVX2 I422TOARGBROW_AVX2 I422TORGB24ROW_AVX2 I422TORGBAROW_AVX2 in row.h.
SSSE3 versions will be used instead.
Short term fix until issue can be resolved.
R=harryjin@google.com
BUG=libyuv:522
Review URL: https://codereview.chromium.org/1419513009 .
On Arm the YVU to RGB conversions move constants into registers.
This change does the same for 64 bit intel builds where additional
registers are available.
The AVX2 saves 3 instructions by because the 2nd argument needs to be a register, so a vmovdqu was avoided.
x64 builds using memory:
AVX2 I420ToARGB_Opt (3059 ms)
SSSE3 I420ToARGB_Opt (3959 ms)
Now using registers
AVX2 I420ToARGB_Opt (2906 ms)
SSSE3 I420ToARGB_Opt (3928 ms)
TBR=harryjin@google.com
BUG=libyuv:520
Review URL: https://codereview.chromium.org/1407353010 .
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 .
U contributes to B and G. V contributes to R and G.
By swapping U and V, they contribute to the opposite channels. Adjust the matrix so the U contribution is in the matrix location such that it till contribute to the
new B channel and vice versa.
This allows ABGR versions of YUV conversion to use the same low level code as ARGB, just using a different matrix and swapping U and V pointers.
As a result the existing I444ToABGRRow functions are no longer needed and are removed.
Previously this function was only Intel AVX2 optimized for Windwos. Now it is also optimized for Arm and GCC.
ARMv7 Neon
Was LibYUVConvertTest.I444ToABGR_Opt (75971 ms)
Now LibYUVConvertTest.I444ToABGR_Opt (3672 ms)
20.6 times faster.
R=xhwang@chromium.org
BUG=libyuv:515
Review URL: https://codereview.chromium.org/1414133006 .
The any function for handling ARGBToI411 was not handling the pixel
replication correctly. On 422 and odd width was handled by duplicating
a pixel of source. 411 needs replication for remainders of 1, 2 or 3
pixels.
The C version was handling odd width but with an average of the remainder
pixels, which does not match the SIMD 'any' handling off remainder.
This changes the odd width handling to mimic the any version.
TBR=harryjin@google.com
BUG=libyuv:491
Review URL: https://codereview.chromium.org/1411733004 .
Allows us to ignore flags passed on to us by Chromium build bots
without having to explicitly disable them. (Thanks pbos!)
TESTED=webrtc ran modules_unittests with a bogus flag did not result in an
error.
R=kjellander@chromium.org
BUG=libyuv:507
Review URL: https://codereview.chromium.org/1417573002 .
This reverts commit 2dd3d9230ee663e71ed4ad9164033ed672e571de.
Reason: chromium_git is a missing variable, and to properly
add gflags, we need to check in GYP files in third_party/gflags
first, then add the DEPS entry.
BUG=libyuv:507
TBR=fbarchard@chromium.org
Review URL: https://codereview.chromium.org/1406323002 .
A hang in color conversion on arm occurs somewhere in yuv to rgb.
Breaking the color test into its own category of test will help
run selective tests to narrow down the issue.
R=harryjin@google.com
BUG=libyuv:506
Review URL: https://codereview.chromium.org/1405543003 .
previously the neon source code was broken into a separate
library built with -mfpu=neon for the neon assembly, while
the C code was built without neon.
In this change, the neon code is added to the main library
and all code built with neon.
TBR=harryjin@google.com
BUG=libyuv:371
Review URL: https://codereview.chromium.org/1392043003 .
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 .
SETUP provided by zhongwei.yao@linaro.org
Previously the 64 bit Neon code had hard coded constants in the setup macro
for YUV conversion, while 32 bit Neon code supported the yuvconstants
parameter.
This change accepts the constants passed to the YUV conversion row function,
allowing different color spaces to be respected - naming JPEG and BT.709.
As well as the existing BT.601.
TBR=harryjin@google.com
BUG=libyuv:472
Review URL: https://codereview.chromium.org/1384323002 .
J444 is JPeg YUV color space with 444 subsampling.
This implementation uses the existing I444ToARGB conversion, which is
BT.601 color space with 444 subsampling, but passing in the jpeg
color matrix constants.
TBR=harryjin@google.com
BUG=449
Review URL: https://codereview.chromium.org/1387313002 .
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 .
API change - I420AlphaToARGB takes flag indicating if RGB should be
premultiplied by alpha.
This version implements an efficient SSSE3 version for Windows.
C version done in 2 steps.
Was
libyuvTest.I420AlphaToARGB_Any (1136 ms)
libyuvTest.I420AlphaToARGB_Unaligned (1210 ms)
libyuvTest.I420AlphaToARGB_Invert (966 ms)
libyuvTest.I420AlphaToARGB_Opt (1031 ms)
libyuvTest.I420AlphaToABGR_Any (1020 ms)
libyuvTest.I420AlphaToABGR_Unaligned (1359 ms)
libyuvTest.I420AlphaToABGR_Invert (1082 ms)
libyuvTest.I420AlphaToABGR_Opt (986 ms)
R=harryjin@google.com
BUG=libyuv:496
Review URL: https://codereview.chromium.org/1367093002 .
Reimplements I444ToARGB as a matrix function.
new I444ToABGR as matrix functions with wrappers and any functions.
Allows for future J444 and H444 versions.
I444ToABGR user level function added.
BUG=libyuv:490, libyuv:449
R=harryjin@google.com
Review URL: https://codereview.chromium.org/1355733002 .
clangcl use compare_win for 32 bit, allowing fallback and enabling avx2 code for clang.
move defines/protos to compare_row.h
fix issue with odd width ARGBCopyAlpha functions by copying destination to temp buffer, then doing alpha copy, then copy back to destination.
R=harryjin@google.comTBR=harryjin@google.com
BUG=libyuv:484
Review URL: https://webrtc-codereview.appspot.com/59379004.
