[Libva] [PATCH intel-driver 0/5] JPEG Encode Test Speed Optimization
Xiang, Haihao
haihao.xiang at intel.com
Tue Nov 22 04:36:48 UTC 2016
Hi Sean,
The configuration is very simple and I ran the test within a text
console. Maybe it is a random issue so it is hard to reproduce in your
side. I will try to debug it.
Thanks
Haihao
> Hello Haihao,
>
> If you like, please file a bug and provide details about your system
> configuration and the errors you are seeing. I'm not able to
> duplicate your errors.
>
> Sean
>
>
> On Sun, Nov 20, 2016 at 7:18 PM, Xiang, Haihao <haihao.xiang at intel.co
> m> wrote:
> >
> > I ran gtest again and got 2 failed cases now:
> >
> > [==========] 580 tests from 16 test cases ran. (111876 ms total)
> > [ PASSED ] 578 tests.
> > [ FAILED ] 2 tests, listed below:
> > [ FAILED ] Common/JPEGEncodeInputTest.Full/97, where GetParam() =
> > (Fixed Size 7680x4320, 0x525ce4 pointing to "UYVY")
> > [ FAILED ] Common/JPEGEncodeInputTest.Full/98, where GetParam() =
> > (Fixed Size 7680x4320, 0x525d10 pointing to "YUY2")
> >
> > however Common/JPEGEncodeInputTest.Full/97 and
> > Common/JPEGEncodeInputTest.Full/98 pass when run the tests
> > separately.
> >
> > [ RUN ] Common/JPEGEncodeInputTest.Full/97
> > Creator: Fixed Size 7680x4320
> > Input : UYVY 7680x4320 {15360,0,0} {4320,0,0} {66355200,0,0}
> > {0,0,0}
> > [ OK ] Common/JPEGEncodeInputTest.Full/97 (6458 ms)
> > [----------] 1 test from Common/JPEGEncodeInputTest (6458 ms total)
> >
> > [ RUN ] Common/JPEGEncodeInputTest.Full/98
> > Creator: Fixed Size 7680x4320
> > Input : YUY2 7680x4320 {15360,0,0} {4320,0,0} {66355200,0,0}
> > {0,0,0}
> > [ OK ] Common/JPEGEncodeInputTest.Full/98 (6235 ms)
> > [----------] 1 test from Common/JPEGEncodeInputTest (6235 ms total)
> >
> > so I guess some buffers/parameters are not cleared before running a
> > new
> > test in a full testing run. I will file a bug to track this issue.
> >
> > Thanks
> > Haihao
> >
> >
> > > Hi Haihao,
> > >
> > > Please file a bug if this persists.
> > >
> > > Thanks,
> > > U. Artie
> > >
> > > > -----Original Message-----
> > > > From: Xiang, Haihao
> > > > Sent: Wednesday, November 09, 2016 7:57 PM
> > > > To: libva at lists.freedesktop.org; seanvk at posteo.de; Eoff,
> > > > Ullysses A
> > > > <ullysses.a.eoff at intel.com>
> > > > Subject: Re: [Libva] [PATCH intel-driver 0/5] JPEG Encode Test
> > > > Speed Optimization
> > > >
> > > >
> > > > Hi Artie,
> > > >
> > > > Did you run gtest on SKL? I saw a failure with this
> > > > optimization
> > > > on
> > > > SKL. After reverting back to commit 63b98a5, the issue
> > > > disappears.
> > > >
> > > > [ RUN ] Common/JPEGEncodeInputTest.Full/97
> > > > i965_jpeg_encode_test.cpp:371: Failure
> > > > Value of: std::abs(ebytes - rbytes).max() <= 2
> > > > Actual: false
> > > > Expected: true
> > > > plane 0: 29641916 of 33177600 (89.3432%) mismatch
> > > > plane 1: 184960 of 16588800 (1.11497%) mismatch
> > > > plane 2: 0 of 16588800 (0%) mismatch
> > > > Creator: Fixed Size 7680x4320
> > > > Input : UYVY 7680x4320 {15360,0,0} {4320,0,0} {66355200,0,0}
> > > > {0,0,0}
> > > > [ FAILED ] Common/JPEGEncodeInputTest.Full/97, where
> > > > GetParam() =
> > > > (Fixed Size 7680x4320, 0x522e64 pointing to "UYVY") (7295 ms)
> > > >
> > > > Thanks
> > > > Haihao
> > > >
> > > >
> > > >
> > > >
> > > > > On Wed, 2016-10-26 at 13:24 -0700, U. Artie Eoff wrote:
> > > > > > This patch series splits out the general YUV input data
> > > > > > fields
> > > > > > and
> > > > > > routines
> > > > > > from the ::JPEG::Encode::TestInput class into it's own
> > > > > > class. This
> > > > > > new class,
> > > > > > YUVImage, was then refactored to use
> > > > > > std::valarray/std::slice
> > > > > > to
> > > > > > manage the
> > > > > > underlying YUV byte data. The JPEG encode tests are
> > > > > > updated
> > > > > > accordingly. This
> > > > > > change allows for faster comparison and conversion
> > > > > > operations
> > > > > > on
> > > > > > YUV
> > > > > > byte data
> > > > > > input/output results and helps reduce some of the copying
> > > > > > overhead.
> > > > > >
> > > > > > Also changed the random value generator to improve it's
> > > > > > speed.
> > > > > >
> > > > > > With these changes, I observe ~2x improvement in runtime
> > > > > > speed
> > > > > > on
> > > > > > the
> > > > > > JPEG
> > > > > > Encode test cases.
> > > > > >
> > > > > > Finally, included is a Timer class for convenience that can
> > > > > > be
> > > > > > used
> > > > > > for future
> > > > > > test development.
> > > > >
> > > > >
> > > > > Nice refactoring that will make it easier for adding the
> > > > > other
> > > > > encoders.
> > > > >
> > > > > lgtm (also tested0, applied.
> > > > >
> > > > > Thanks,
> > > > >
> > > > > Sean
> > > > >
> > > > > >
> > > > > > U. Artie Eoff (5):
> > > > > > test: add a timer class
> > > > > > test: use C random library for random numbers
> > > > > > test: add YUVImage class
> > > > > > test: streamable valarray
> > > > > > test: use YUVImage in JPEG encode tests
> > > > > >
> > > > > > test/Makefile.am | 2 +
> > > > > > test/i965_jpeg_encode_test.cpp | 135 +++-----
> > > > > > test/i965_jpeg_test_data.cpp | 196 ++---------
> > > > > > test/i965_jpeg_test_data.h | 26 +-
> > > > > > test/i965_streamable.h | 25 ++
> > > > > > test/i965_test_environment.cpp | 7 +
> > > > > > test/i965_test_image_utils.cpp | 747
> > > > > > +++++++++++++++++++++++++++++++++++++++++
> > > > > > test/i965_test_image_utils.h | 66 ++++
> > > > > > test/object_heap_test.cpp | 4 -
> > > > > > test/test_utils.h | 45 ++-
> > > > > > 10 files changed, 970 insertions(+), 283 deletions(-)
> > > > > > create mode 100644 test/i965_test_image_utils.cpp
> > > > > > create mode 100644 test/i965_test_image_utils.h
> > > > > >
> > > > > _______________________________________________
> > > > > Libva mailing list
> > > > > Libva at lists.freedesktop.org
> > > > > https://lists.freedesktop.org/mailman/listinfo/libva
> > _______________________________________________
> > Libva mailing list
> > Libva at lists.freedesktop.org
> > https://lists.freedesktop.org/mailman/listinfo/libva
>
>
>
More information about the Libva
mailing list