add ConvertYUVAToARGBRow_MMX to media_perftests

Add YUVConvertPerfTest.ConvertYUVAToARGBRow_MMX and YUVConvertPerfTest.I422ToARGBRow_SSSE3 to media_perftests.
Fix math error when printing results by clearing the MMX state before doing float math.
BUG=libyuv:473
Committed: https://crrev.com/98e918a3abdfcc2eaae3df9cc3dd9f8225f39edc
Cr-Commit-Position: refs/heads/master@{#343140}

[fbarchard, 2015-08-12 19:00:59]

fbarchard@chromium.org changed reviewers:
+ dalecurtis@chromium.org, fbarchard@chromium.org

[fbarchard, 2015-08-12 19:01:00]

Added benchmarks for yuva and libyuv

[ RUN      ] YUVConvertPerfTest.I422ToARGBRow_SSSE3
*RESULT yuv_convert_perftest: I422ToARGBRow_SSSE3= 8529.111991505004 runs/s
[       OK ] YUVConvertPerfTest.I422ToARGBRow_SSSE3 (235 ms)
[ RUN      ] YUVConvertPerfTest.ConvertYUVToRGB32Row_SSE
*RESULT yuv_convert_perftest: ConvertYUVToRGB32Row_SSE= -NaN runs/s
[       OK ] YUVConvertPerfTest.ConvertYUVToRGB32Row_SSE (562 ms)
[ RUN      ] YUVConvertPerfTest.ConvertYUVAToARGBRow_MMX
*RESULT yuv_convert_perftest: ConvertYUVAToARGBRow_MMX= -NaN runs/s
[       OK ] YUVConvertPerfTest.ConvertYUVAToARGBRow_MMX (852 ms)

[brucedawson, 2015-08-12 19:07:15]

-NaN runs/s seems dodgy. The other results should be printed with fewer than 16
digits of precision as it makes the results harder to read.

[DaleCurtis, 2015-08-12 21:03:12]

Agree w/ Bruce, are the start and end times the same or something on those
tests? Seems hard to believe.

[fbarchard1, 2015-08-12 23:20:58]

full path for includes

[fbarchard, 2015-08-12 23:27:05]

Done.

[----------] 5 tests from YUVConvertPerfTest
[ RUN      ] YUVConvertPerfTest.I422ToARGBRow_SSSE3
*RESULT yuv_convert_perftest: I422ToARGBRow_SSSE3= 8699.585899711174 runs/s
[       OK ] YUVConvertPerfTest.I422ToARGBRow_SSSE3 (231 ms)
[ RUN      ] YUVConvertPerfTest.ConvertYUVToRGB32Row_SSE
*RESULT yuv_convert_perftest: ConvertYUVToRGB32Row_SSE= 3527.504837091008 runs/s
[       OK ] YUVConvertPerfTest.ConvertYUVToRGB32Row_SSE (568 ms)
[ RUN      ] YUVConvertPerfTest.ConvertYUVAToARGBRow_MMX
*RESULT yuv_convert_perftest: ConvertYUVAToARGBRow_MMX= 2432.7999824838403
runs/s
[       OK ] YUVConvertPerfTest.ConvertYUVAToARGBRow_MMX (823 ms)
[ RUN      ] YUVConvertPerfTest.ScaleYUVToRGB32Row_SSE
*RESULT yuv_convert_perftest: ScaleYUVToRGB32Row_SSE= 2654.1320855373688 runs/s
[       OK ] YUVConvertPerfTest.ScaleYUVToRGB32Row_SSE (754 ms)
[ RUN      ] YUVConvertPerfTest.LinearScaleYUVToRGB32Row_SSE
*RESULT yuv_convert_perftest: LinearScaleYUVToRGB32Row_SSE= 1401.0861219617448
runs/s
[       OK ] YUVConvertPerfTest.LinearScaleYUVToRGB32Row_SSE (1429 ms)
[----------] 5 tests from YUVConvertPerfTest (3825 ms total)

https://codereview.chromium.org/1286913002/diff/40001/media/base/yuv_convert_...
File media/base/yuv_convert_perftest.cc (right):

https://codereview.chromium.org/1286913002/diff/40001/media/base/yuv_convert_...
media/base/yuv_convert_perftest.cc:69: ASSERT_TRUE(base::CPU().has_ssse3());
Is assert ok?

https://codereview.chromium.org/1286913002/diff/40001/media/base/yuv_convert_...
media/base/yuv_convert_perftest.cc:129: media::EmptyRegisterState();
media::EmptyRegisterState() does emms, which allows the float math that follows
to work.

[fbarchard, 2015-08-13 00:43:18]

Tested using a Pentium4 emulator:
bin32\drrun -t drcpusim -continue -cpu Prescott -fool_cpuid --
out\Release\media_perftests.exe
--single-process-tests --gtest_catch_exceptions=0 --gtest_filter=*Conv*

[ RUN      ] YUVConvertPerfTest.I422ToARGBRow_SSSE3
c:\src\chrome\src\media\base\yuv_convert_perftest.cc(69): error: Value of:
base::CPU().has_ssse3()
  Actual: false
Expected: true
[  FAILED  ] YUVConvertPerfTest.I422ToARGBRow_SSSE3 (17 ms)
...
[  PASSED  ] 7 tests.
[  FAILED  ] 1 test, listed below:
[  FAILED  ] YUVConvertPerfTest.I422ToARGBRow_SSSE3

[DaleCurtis, 2015-08-13 00:46:40]

Thanks for fixing the MMX issue, lgtm

https://codereview.chromium.org/1286913002/diff/40001/media/base/yuv_convert_...
File media/base/yuv_convert_perftest.cc (right):

https://codereview.chromium.org/1286913002/diff/40001/media/base/yuv_convert_...
media/base/yuv_convert_perftest.cc:69: ASSERT_TRUE(base::CPU().has_ssse3());
On 2015/08/12 23:27:05, fbarchard wrote:
> Is assert ok?

If the bots pass, then yes. It'll be obvious if they don't :)

[fbarchard, 2015-08-13 01:04:58]

The CQ bit was checked by fbarchard@chromium.org

[commit-bot: I haz the power, 2015-08-13 01:05:12]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1286913002/40001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1286913002/40001

[commit-bot: I haz the power, 2015-08-13 01:17:16]

Committed patchset #3 (id:40001)

[commit-bot: I haz the power, 2015-08-13 01:17:50]

Patchset 3 (id:??) landed as
https://crrev.com/98e918a3abdfcc2eaae3df9cc3dd9f8225f39edc
Cr-Commit-Position: refs/heads/master@{#343140}

[Ken Rockot(use gerrit already), 2015-08-13 01:28:21]

A revert of this CL (patchset #3 id:40001) has been created in
https://codereview.chromium.org/1288223003/ by rockot@chromium.org.

The reason for reverting is: This broke the build:

http://build.chromium.org/p/chromium.linux/buildstatus?builder=Linux%20GN%20C...

missing libyuv/basic_types.h

.

[DaleCurtis, 2015-08-13 16:24:45]

Whoops, yeah you need to update the BUILD.gn file as well.