The sum of floats can optimize differently with vectorization, producing
a different result between NEON and C.
Adjust the unittest to allow for some difference in the sum.
The NEON version is 8 samples at a time, so the test now rounds up
the number of values to multiple of 8.
TBR=kjellander@chromium.org
Bug: libyuv:717
Test: LibYUVPlanarTest.TestScaleSumSamples_Opt
Change-Id: I2a0783780c7e0f240f7a8e4700b2a4d3e6b52d87
Reviewed-on: https://chromium-review.googlesource.com/673708
Reviewed-by: Cheng Wang <wangcheng@google.com>
When command line --libyuv_cpu_info is used the individual tests
used to need to set the cpumask. This CL moves that to the init
for each test class so the individual tests dont need to set it.
TBR=kjellander@chromium.org
BUG=libyuv:720
TEST=LibYUVBaseTest.TestCpuHas
Change-Id: I6ae180388debf6cf76be6df5b81cfffeb35ee2eb
Reviewed-on: https://chromium-review.googlesource.com/662367
Reviewed-by: Cheng Wang <wangcheng@google.com>
Reduce buffers for test to 640 from 1280 to avoid
bit stack warning.
TBR=kjellander@chromium.org
BUG=libyuv:730
TEST=LibYUVPlanarTest.TestGaussRow_Opt and LibYUVPlanarTest.TestGaussCol_Opt
Change-Id: I710af3e952f9a4d1c0c0c8f73922c1d98ad9aa29
Reviewed-on: https://chromium-review.googlesource.com/660662
Reviewed-by: Frank Barchard <fbarchard@google.com>
Roughly. instead of 4 loads and 8 multiples, use 1 load and 2 multiples
4 times over. The original code, as with the C code from clang and gcc,
did all the loads, then all the math, then the store. The new code
does a load, then the math, then the next load, etc.
This schedules better on current arm 64 cpus.
Number of registers also reduced, reusing the same registers.
HiSilicon ARM A73:
Now
TestGaussRow_Opt (890 ms)
TestGaussCol_Opt (571 ms)
Was
TestGaussRow_Opt (1061 ms)
TestGaussCol_Opt (595 ms)
Qualcomm 821 (Pixel):
Now
TestGaussRow_Opt (571 ms)
TestGaussCol_Opt (474 ms)
Was
TestGaussRow_Opt (751 ms)
TestGaussCol_Opt (520 ms)
TBR=kjellander@chromium.org
BUG=libyuv:719
TEST=LibYUVPlanarTest.TestGaussRow_Opt
Reviewed-on: https://chromium-review.googlesource.com/627478
Reviewed-by: Cheng Wang <wangcheng@google.com>
Reviewed-by: Frank Barchard <fbarchard@google.com>
Change-Id: I5ec81191d460801f0d4a89f0384f89925ff036de
Reviewed-on: https://chromium-review.googlesource.com/634448
Commit-Queue: Frank Barchard <fbarchard@google.com>
add ScaleMaxSamples_NEON function with max
done on original values.
TBR=kjellander@chromium.org
BUG=libyuv:717
TEST=LibYUVPlanarTest.TestScaleMaxSamples_Opt
Change-Id: Id99338860782b10ffd24f66242eb42014c2e229e
Reviewed-on: https://chromium-review.googlesource.com/614685
Reviewed-by: Frank Barchard <fbarchard@google.com>
Reviewed-by: Cheng Wang <wangcheng@google.com>
BUG=None
TEST=try bots and lint test
Change-Id: I1ab462adf2d309117862c5eb4b244a61ae202951
Reviewed-on: https://chromium-review.googlesource.com/450658
Commit-Queue: Frank Barchard <fbarchard@google.com>
Reviewed-by: Henrik Kjellander <kjellander@chromium.org>
android.mk builds have unused parameter warning on by default.
This change for GN makes libyuv build the same way.
BUG=libyuv:681
TEST=build on linux with clang and ninja.
Change-Id: I76c627d446b96653f147725bca915d94a42ce9a6
Reviewed-on: https://chromium-review.googlesource.com/441194
Reviewed-by: Frank Barchard <fbarchard@google.com>
Commit-Queue: Frank Barchard <fbarchard@google.com>
Halffloats have a limited range. It shouldnt normally come up, but if the scale value passed in produces a small value, the half floats will be denormals, which are slow and/or flust to zero. This test ensures they behave the same in C and SIMD and tests the performance of denormals.
TEST=TestHalfFloatPlane_denormal
BUG=libyuv:560
R=hubbe@chromium.org
Review URL: https://codereview.chromium.org/2424233004 .
Add test for SplitUVPlane and MergeUVPlane
Add public methods SplitUVPlanes and MergeUVPlanes based on the
optimized assembly functions that already exists.
TEST=SplitUVPlane unittest
BUG=libyuv:629
R=braveyao@chromium.org
Review URL: https://codereview.chromium.org/2279603002 .
blur requires memory be aligned. change the unittest allocator to guarantee 64 byte alignment.
re-enable blur any test that fails if memory is unaligned.
TBR=harryjin@google.com
BUG=libyuv:596,libyuv:594
TESTED=local build passes with row.h removed from tests.
Review URL: https://codereview.chromium.org/2019753002 .
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 .
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 .
in order to compare C and Neon code, a new command line flag is added.
historically environment variables controlled cpu features, but on
android apk it is easier to pass a command line option to disable cpu
optimizations.
R=harryjin@google.com
BUG=libyuv:516
Review URL: https://codereview.chromium.org/1407193009 .
random / rand is slow and impacts performance testing.
Although its only called to clear a frame once, a typical profile shows
it high in the overall profile, when doing 1000 frames for a benchmark.
95.10% libyuv_unittest libyuv_unittest [.] YUY2ToARGBRow_SSSE3
2.01% libyuv_unittest libc-2.19.so [.] __random_r
1.13% libyuv_unittest libc-2.19.so [.] __random
Replace random is a faster version for unittests.
set LIBYUV_WIDTH=1280
set LIBYUV_HEIGHT=720
set LIBYUV_REPEAT=999
set LIBYUV_FLAGS=-1
out\release\libyuv_unittest --gtest_filter=*YUY2ToARGB* | findms
Was
libyuvTest.YUY2ToARGB_Opt (497 ms)
Now
libyuvTest.YUY2ToARGB_Opt (454 ms)
R=harryjin@google.com
BUG=none
Review URL: https://codereview.chromium.org/1361813002 .
