Re: [Libva] [Libva-intel-driver][PATCH] Use Media Read message if possible on Gen8+

2016-10-18 Thread Eoff, Ullysses A 
I've created a bugzilla for the issue... 
https://bugs.freedesktop.org/show_bug.cgi?id=98311

Let's continue discussion there.

U. Artie

> -Original Message-
> From: Sean V Kelley [mailto:sean.v.kel...@intel.com]
> Sent: Monday, October 17, 2016 10:35 AM
> To: Xiang, Haihao <haihao.xi...@intel.com>
> Cc: Zhao, Yakui <yakui.z...@intel.com>; Eoff, Ullysses A 
> <ullysses.a.e...@intel.com>; libva@lists.freedesktop.org
> Subject: Re: [Libva] [Libva-intel-driver][PATCH] Use Media Read message if 
> possible on Gen8+
> 
> Yes, I want to throttle the pace of the changes until we qualify and
> understand the performance and test failure issues. Then we can resume
> normal flow.
> 
> For these tests to be meaningful they have to be consistent across
> distributions and platforms.
> 
> Thanks,
> 
> Sean
> 
> 
> On Sun, Oct 16, 2016 at 8:11 PM, Xiang, Haihao <haihao.xi...@intel.com> wrote:
> >
> > Hi Sean,
> >
> > I can't reproduce the issue mentioned by you and Artie on my SKL. I am
> > using Ubuntu 14.04 but built kernel and libdrm from source code:
> >
> > Linux kernel:
> > aab15c274da587bcab19376d2caa9d6626440335 (drm-intel-nightly: 2016y-09m-
> > 26d-12h-11m-33s)
> >
> > libdrm: 2.4.70
> >
> > I don't think the bus error is related to this patch, but I would hold
> > off on merging this patch if you reproduced the error with
> > Common/JPEGEncodeInputTest.Full/95 only.
> >
> > Thanks
> > Haihao
> >
> >
> >
> >> On 10/15/2016 03:58 AM, Eoff, Ullysses A wrote:
> >> >
> >> > > -Original Message-
> >> > > From: Libva [mailto:libva-boun...@lists.freedesktop.org] On
> >> > > Behalf Of Sean V Kelley
> >> > > Sent: Friday, October 14, 2016 11:44 AM
> >> > > To: Zhao, Yakui<yakui.z...@intel.com>; Xiang, Haihao >> > > @intel.com>
> >> > > Cc: libva@lists.freedesktop.org
> >> > > Subject: Re: [Libva] [Libva-intel-driver][PATCH] Use Media Read
> >> > > message if possible on Gen8+
> >> > >
> >> > > On Fri, 2016-10-14 at 08:37 +0800, Zhao Yakui wrote:
> >> > > > On 10/13/2016 03:07 AM, Xiang, Haihao wrote:
> >> > > > > AVS can't gurantee bit-match for a large surface. This fixes
> >> > > > > the
> >> > > > > failure reported by gtest case
> >> > > > > Common/JPEGEncodeInputTest.Full/95.
> >> > > > >
> >> > > > > before:
> >> > > > > [  FAILED  ] Common/JPEGEncodeInputTest.Full/95, where
> >> > > > > GetParam() =
> >> > > > > (Fixed Size 7680x4320, 0x501176 pointing to "I420") (9239 ms)
> >> > > > > [--] 1 test from Common/JPEGEncodeInputTest (9239 ms
> >> > > > > total)
> >> > > > >
> >> > > > > [--] Global test environment tear-down
> >> > > > > [==] 1 test from 1 test case ran. (9361 ms total)
> >> > > > > [  PASSED  ] 0 tests.
> >> > > > > [  FAILED  ] 1 test, listed below:
> >> > > > > [  FAILED  ] Common/JPEGEncodeInputTest.Full/95, where
> >> > > > > GetParam() =
> >> > > > > (Fixed Size 7680x4320, 0x501176 pointing to "I420")
> >> > > > >
> >> > > > > after:
> >> > > > > [   OK ] Common/JPEGEncodeInputTest.Full/95 (15250 ms)
> >> > > > > [--] 1 test from Common/JPEGEncodeInputTest (15250 ms
> >> > > > > total)
> >> > > > >
> >> > > > > [--] Global test environment tear-down
> >> > > > > [==] 1 test from 1 test case ran. (15365 ms total)
> >> > > > > [  PASSED  ] 1 test.
> >> > > > >
> >> > > > > Signed-off-by: Xiang, Haihao<haihao.xi...@intel.com>
> >> > > >
> >> > > > This looks good to me.
> >> > > >
> >> > > > Add: Reviewed-by: Zhao Yakui<yakui.z...@intel.com>
> >> > >
> >> > >
> >> > > Getting a core dump although the test is now passing.  We'll need
> >> > > to
> >> > > debug further.
> >> > >
> >> >
> >> > I don't get any core dump on my SKL...  And the test passes for me
> >>

Re: [Libva] [Libva-intel-driver][PATCH] Use Media Read message if possible on Gen8+

2016-10-17 Thread Sean V Kelley 
Yes, I want to throttle the pace of the changes until we qualify and
understand the performance and test failure issues. Then we can resume
normal flow.

For these tests to be meaningful they have to be consistent across
distributions and platforms.

Thanks,

Sean


On Sun, Oct 16, 2016 at 8:11 PM, Xiang, Haihao <haihao.xi...@intel.com> wrote:
>
> Hi Sean,
>
> I can't reproduce the issue mentioned by you and Artie on my SKL. I am
> using Ubuntu 14.04 but built kernel and libdrm from source code:
>
> Linux kernel:
> aab15c274da587bcab19376d2caa9d6626440335 (drm-intel-nightly: 2016y-09m-
> 26d-12h-11m-33s)
>
> libdrm: 2.4.70
>
> I don't think the bus error is related to this patch, but I would hold
> off on merging this patch if you reproduced the error with
> Common/JPEGEncodeInputTest.Full/95 only.
>
> Thanks
> Haihao
>
>
>
>> On 10/15/2016 03:58 AM, Eoff, Ullysses A wrote:
>> >
>> > > -Original Message-
>> > > From: Libva [mailto:libva-boun...@lists.freedesktop.org] On
>> > > Behalf Of Sean V Kelley
>> > > Sent: Friday, October 14, 2016 11:44 AM
>> > > To: Zhao, Yakui<yakui.z...@intel.com>; Xiang, Haihao> > > @intel.com>
>> > > Cc: libva@lists.freedesktop.org
>> > > Subject: Re: [Libva] [Libva-intel-driver][PATCH] Use Media Read
>> > > message if possible on Gen8+
>> > >
>> > > On Fri, 2016-10-14 at 08:37 +0800, Zhao Yakui wrote:
>> > > > On 10/13/2016 03:07 AM, Xiang, Haihao wrote:
>> > > > > AVS can't gurantee bit-match for a large surface. This fixes
>> > > > > the
>> > > > > failure reported by gtest case
>> > > > > Common/JPEGEncodeInputTest.Full/95.
>> > > > >
>> > > > > before:
>> > > > > [  FAILED  ] Common/JPEGEncodeInputTest.Full/95, where
>> > > > > GetParam() =
>> > > > > (Fixed Size 7680x4320, 0x501176 pointing to "I420") (9239 ms)
>> > > > > [--] 1 test from Common/JPEGEncodeInputTest (9239 ms
>> > > > > total)
>> > > > >
>> > > > > [--] Global test environment tear-down
>> > > > > [==] 1 test from 1 test case ran. (9361 ms total)
>> > > > > [  PASSED  ] 0 tests.
>> > > > > [  FAILED  ] 1 test, listed below:
>> > > > > [  FAILED  ] Common/JPEGEncodeInputTest.Full/95, where
>> > > > > GetParam() =
>> > > > > (Fixed Size 7680x4320, 0x501176 pointing to "I420")
>> > > > >
>> > > > > after:
>> > > > > [   OK ] Common/JPEGEncodeInputTest.Full/95 (15250 ms)
>> > > > > [--] 1 test from Common/JPEGEncodeInputTest (15250 ms
>> > > > > total)
>> > > > >
>> > > > > [--] Global test environment tear-down
>> > > > > [==] 1 test from 1 test case ran. (15365 ms total)
>> > > > > [  PASSED  ] 1 test.
>> > > > >
>> > > > > Signed-off-by: Xiang, Haihao<haihao.xi...@intel.com>
>> > > >
>> > > > This looks good to me.
>> > > >
>> > > > Add: Reviewed-by: Zhao Yakui<yakui.z...@intel.com>
>> > >
>> > >
>> > > Getting a core dump although the test is now passing.  We'll need
>> > > to
>> > > debug further.
>> > >
>> >
>> > I don't get any core dump on my SKL...  And the test passes for me
>> > with
>> > this patch.  Are you running the test in isolation or with the
>> > entire suite?
>> > If with the entire suite, does the core dump occur during this test
>> > case or
>> > another?
>> >
>> > I've noticed that I occasionally get a "bus error" with the
>> > Big/JPEGEncodeInputTest.* test cases (i.e. 8192x8192) even without
>> > this patch.  Perhaps this is what you're encountering?
>>
>> Hi, Sean/Artie
>>
>>  Is it possible that you can send out the dmesg log when the
>> bus_error is triggered?
>>  I try the test several times on one KBL machine(similar to SKL)
>> and
>> unfortunately there is no "sig_bus error".
>>
>> Thanks
>> Yakui
>>
>> >
>> > U. Artie
>> >
>> > > Sean
>> > >
>> > > >
>> > > > Thanks
>> > > > Yakui
>&g

Re: [Libva] [Libva-intel-driver][PATCH] Use Media Read message if possible on Gen8+

2016-10-16 Thread Xiang, Haihao 

> On Fri, 2016-10-14 at 19:58 +, Eoff, Ullysses A wrote:
> > > -Original Message-
> > > From: Libva [mailto:libva-boun...@lists.freedesktop.org] On
> > > Behalf
> > > Of Sean V Kelley
> > > Sent: Friday, October 14, 2016 11:44 AM
> > > To: Zhao, Yakui <yakui.z...@intel.com>; Xiang, Haihao
> > >  > > @intel.com>
> > > Cc: libva@lists.freedesktop.org
> > > Subject: Re: [Libva] [Libva-intel-driver][PATCH] Use Media Read
> > > message if possible on Gen8+
> > > 
> > > On Fri, 2016-10-14 at 08:37 +0800, Zhao Yakui wrote:
> > > > On 10/13/2016 03:07 AM, Xiang, Haihao wrote:
> > > > > AVS can't gurantee bit-match for a large surface. This fixes
> > > > > the
> > > > > failure reported by gtest case
> > > > > Common/JPEGEncodeInputTest.Full/95.
> > > > > 
> > > > > before:
> > > > > [  FAILED  ] Common/JPEGEncodeInputTest.Full/95, where
> > > > > GetParam() =
> > > > > (Fixed Size 7680x4320, 0x501176 pointing to "I420") (9239 ms)
> > > > > [--] 1 test from Common/JPEGEncodeInputTest (9239 ms
> > > > > total)
> > > > > 
> > > > > [--] Global test environment tear-down
> > > > > [==] 1 test from 1 test case ran. (9361 ms total)
> > > > > [  PASSED  ] 0 tests.
> > > > > [  FAILED  ] 1 test, listed below:
> > > > > [  FAILED  ] Common/JPEGEncodeInputTest.Full/95, where
> > > > > GetParam() =
> > > > > (Fixed Size 7680x4320, 0x501176 pointing to "I420")
> > > > > 
> > > > > after:
> > > > > [   OK ] Common/JPEGEncodeInputTest.Full/95 (15250 ms)
> > > > > [--] 1 test from Common/JPEGEncodeInputTest (15250 ms
> > > > > total)
> > > > > 
> > > > > [--] Global test environment tear-down
> > > > > [==] 1 test from 1 test case ran. (15365 ms total)
> > > > > [  PASSED  ] 1 test.
> > > > > 
> > > > > Signed-off-by: Xiang, Haihao<haihao.xi...@intel.com>
> > > > 
> > > > This looks good to me.
> > > > 
> > > > Add: Reviewed-by: Zhao Yakui <yakui.z...@intel.com>
> > > 
> > > 
> > > Getting a core dump although the test is now passing.  We'll need
> > > to
> > > debug further.
> > > 
> > 
> > I don't get any core dump on my SKL...  And the test passes for me
> > with
> > this patch.  Are you running the test in isolation or with the
> > entire
> > suite?
> > If with the entire suite, does the core dump occur during this test
> > case or
> > another?
> > 
> > I've noticed that I occasionally get a "bus error" with the
> > Big/JPEGEncodeInputTest.* test cases (i.e. 8192x8192) even without
> > this patch.  Perhaps this is what you're encountering?
> 
> I'm consistently seeing this on a SKL system.  It is also quite slow
> for execution.  Actually tested this on my Arch sytem with Kernel 4.7
> and the older LTS 4.4.  Same bus error and slow execution
> performance.

Sorry, I didn't notice the issue of slow execution. In theory, my patch
should not cause slowdown because it read YUV data directly from a
surface in the GPU kernel. I ran the test case several times
with/without the patch, and can't reproduce the issue. It seems this
issue is also a random issue in gtest.

> 
> We need to isolate the root cause here because I suspect this is what
> Haihao has also been seeing.
> 
> Takes nearly a minute on one system and far less time on an Ubuntu
> system with the same kernel and essentially the same SKL SKU.
> 
> Sean
> 
> 
> 
> > 
> > U. Artie
> > 
> > > Sean
> > > 
> > > > 
> > > > Thanks
> > > >    Yakui
> > > > 
> > > > > ---
> > > > >   src/gen8_post_processing.c | 56
> > > > > +-
> > > > >   src/shaders/post_processing/gen8/Makefile.am   |  2 +
> > > > >   .../gen8/PL2_media_read_buf0123.g8a| 65
> > > > > +
> > > > >   .../gen8/PL3_media_read_buf0123.g8a| 68
> > > > > ++
> > > > >   src/shaders/post_processing/gen8/pl2_to_pl2.asm|  2 +
> > > > >   src/shaders/post_pr

Re: [Libva] [Libva-intel-driver][PATCH] Use Media Read message if possible on Gen8+

2016-10-16 Thread Xiang, Haihao 

Hi Sean,

I can't reproduce the issue mentioned by you and Artie on my SKL. I am
using Ubuntu 14.04 but built kernel and libdrm from source code:

Linux kernel:
aab15c274da587bcab19376d2caa9d6626440335 (drm-intel-nightly: 2016y-09m-
26d-12h-11m-33s)

libdrm: 2.4.70

I don't think the bus error is related to this patch, but I would hold
off on merging this patch if you reproduced the error with
Common/JPEGEncodeInputTest.Full/95 only.

Thanks
Haihao



> On 10/15/2016 03:58 AM, Eoff, Ullysses A wrote:
> > 
> > > -Original Message-
> > > From: Libva [mailto:libva-boun...@lists.freedesktop.org] On
> > > Behalf Of Sean V Kelley
> > > Sent: Friday, October 14, 2016 11:44 AM
> > > To: Zhao, Yakui<yakui.z...@intel.com>; Xiang, Haihao > > @intel.com>
> > > Cc: libva@lists.freedesktop.org
> > > Subject: Re: [Libva] [Libva-intel-driver][PATCH] Use Media Read
> > > message if possible on Gen8+
> > > 
> > > On Fri, 2016-10-14 at 08:37 +0800, Zhao Yakui wrote:
> > > > On 10/13/2016 03:07 AM, Xiang, Haihao wrote:
> > > > > AVS can't gurantee bit-match for a large surface. This fixes
> > > > > the
> > > > > failure reported by gtest case
> > > > > Common/JPEGEncodeInputTest.Full/95.
> > > > > 
> > > > > before:
> > > > > [  FAILED  ] Common/JPEGEncodeInputTest.Full/95, where
> > > > > GetParam() =
> > > > > (Fixed Size 7680x4320, 0x501176 pointing to "I420") (9239 ms)
> > > > > [--] 1 test from Common/JPEGEncodeInputTest (9239 ms
> > > > > total)
> > > > > 
> > > > > [--] Global test environment tear-down
> > > > > [==] 1 test from 1 test case ran. (9361 ms total)
> > > > > [  PASSED  ] 0 tests.
> > > > > [  FAILED  ] 1 test, listed below:
> > > > > [  FAILED  ] Common/JPEGEncodeInputTest.Full/95, where
> > > > > GetParam() =
> > > > > (Fixed Size 7680x4320, 0x501176 pointing to "I420")
> > > > > 
> > > > > after:
> > > > > [   OK ] Common/JPEGEncodeInputTest.Full/95 (15250 ms)
> > > > > [--] 1 test from Common/JPEGEncodeInputTest (15250 ms
> > > > > total)
> > > > > 
> > > > > [--] Global test environment tear-down
> > > > > [==] 1 test from 1 test case ran. (15365 ms total)
> > > > > [  PASSED  ] 1 test.
> > > > > 
> > > > > Signed-off-by: Xiang, Haihao<haihao.xi...@intel.com>
> > > > 
> > > > This looks good to me.
> > > > 
> > > > Add: Reviewed-by: Zhao Yakui<yakui.z...@intel.com>
> > > 
> > > 
> > > Getting a core dump although the test is now passing.  We'll need
> > > to
> > > debug further.
> > > 
> > 
> > I don't get any core dump on my SKL...  And the test passes for me
> > with
> > this patch.  Are you running the test in isolation or with the
> > entire suite?
> > If with the entire suite, does the core dump occur during this test
> > case or
> > another?
> > 
> > I've noticed that I occasionally get a "bus error" with the
> > Big/JPEGEncodeInputTest.* test cases (i.e. 8192x8192) even without
> > this patch.  Perhaps this is what you're encountering?
> 
> Hi, Sean/Artie
> 
>  Is it possible that you can send out the dmesg log when the 
> bus_error is triggered?
>  I try the test several times on one KBL machine(similar to SKL)
> and 
> unfortunately there is no "sig_bus error".
> 
> Thanks
> Yakui
> 
> > 
> > U. Artie
> > 
> > > Sean
> > > 
> > > > 
> > > > Thanks
> > > > Yakui
> > > > 
> > > > > ---
> > > > >    src/gen8_post_processing.c | 56
> > > > > +-
> > > > >    src/shaders/post_processing/gen8/Makefile.am   |  2 +
> > > > >    .../gen8/PL2_media_read_buf0123.g8a| 65
> > > > > +
> > > > >    .../gen8/PL3_media_read_buf0123.g8a| 68
> > > > > ++
> > > > >    src/shaders/post_processing/gen8/pl2_to_pl2.asm|  2 +
> > > > >    src/shaders/post_processing/gen8/pl2_to_pl2.g8b| 44
> > > > > ++
> > > > >    sr

Re: [Libva] [Libva-intel-driver][PATCH] Use Media Read message if possible on Gen8+

2016-10-14 Thread Eoff, Ullysses A 




U. Artie Eoff
Software Engineer
Visualization - Intel Open Source Technology Center
Hillsboro, Oregon, USA

> -Original Message-
> From: Kelley, Sean V
> Sent: Friday, October 14, 2016 2:48 PM
> To: Eoff, Ullysses A <ullysses.a.e...@intel.com>; sea...@posteo.de; Zhao, 
> Yakui <yakui.z...@intel.com>; Xiang, Haihao
> <haihao.xi...@intel.com>
> Cc: libva@lists.freedesktop.org
> Subject: Re: [Libva] [Libva-intel-driver][PATCH] Use Media Read message if 
> possible on Gen8+
> 
> 
> 
> On 10/14/16, 2:12 PM, "Libva on behalf of Eoff, Ullysses A" 
> <libva-boun...@lists.freedesktop.org on behalf of
> ullysses.a.e...@intel.com> wrote:
> 
> > -Original Message-
> > From: Sean V Kelley [mailto:sea...@posteo.de]
> > Sent: Friday, October 14, 2016 1:45 PM
> > To: Eoff, Ullysses A <ullysses.a.e...@intel.com>; Zhao, Yakui 
> <yakui.z...@intel.com>; Xiang, Haihao <haihao.xi...@intel.com>
>     > Cc: libva@lists.freedesktop.org
> > Subject: Re: [Libva] [Libva-intel-driver][PATCH] Use Media Read message 
> if possible on Gen8+
> >
> > On Fri, 2016-10-14 at 19:58 +, Eoff, Ullysses A wrote:
> > > > -Original Message-
> > > > From: Libva [mailto:libva-boun...@lists.freedesktop.org] On Behalf
> > > > Of Sean V Kelley
> > > > Sent: Friday, October 14, 2016 11:44 AM
>     > > > To: Zhao, Yakui <yakui.z...@intel.com>; Xiang, Haihao  > > > @intel.com>
> > > > Cc: libva@lists.freedesktop.org
> > > > Subject: Re: [Libva] [Libva-intel-driver][PATCH] Use Media Read
> > > > message if possible on Gen8+
> > > >
> > > > On Fri, 2016-10-14 at 08:37 +0800, Zhao Yakui wrote:
> > > > > On 10/13/2016 03:07 AM, Xiang, Haihao wrote:
> > > > > > AVS can't gurantee bit-match for a large surface. This fixes
> > > > > > the
> > > > > > failure reported by gtest case
> > > > > > Common/JPEGEncodeInputTest.Full/95.
> > > > > >
> > > > > > before:
> > > > > > [  FAILED  ] Common/JPEGEncodeInputTest.Full/95, where
> > > > > > GetParam() =
> > > > > > (Fixed Size 7680x4320, 0x501176 pointing to "I420") (9239 ms)
> > > > > > [--] 1 test from Common/JPEGEncodeInputTest (9239 ms
> > > > > > total)
> > > > > >
> > > > > > [--] Global test environment tear-down
> > > > > > [==] 1 test from 1 test case ran. (9361 ms total)
> > > > > > [  PASSED  ] 0 tests.
> > > > > > [  FAILED  ] 1 test, listed below:
> > > > > > [  FAILED  ] Common/JPEGEncodeInputTest.Full/95, where
> > > > > > GetParam() =
> > > > > > (Fixed Size 7680x4320, 0x501176 pointing to "I420")
> > > > > >
> > > > > > after:
> > > > > > [   OK ] Common/JPEGEncodeInputTest.Full/95 (15250 ms)
> > > > > > [--] 1 test from Common/JPEGEncodeInputTest (15250 ms
> > > > > > total)
> > > > > >
> > > > > > [--] Global test environment tear-down
> > > > > > [==] 1 test from 1 test case ran. (15365 ms total)
> > > > > > [  PASSED  ] 1 test.
> > > > > >
> > > > > > Signed-off-by: Xiang, Haihao<haihao.xi...@intel.com>
> > > > >
> > > > > This looks good to me.
> > > > >
> > > > > Add: Reviewed-by: Zhao Yakui <yakui.z...@intel.com>
> > > >
> > > >
> > > > Getting a core dump although the test is now passing.  We'll need
> > > > to
> > > > debug further.
> > > >
> > >
> > > I don't get any core dump on my SKL...  And the test passes for me
> > > with
> > > this patch.  Are you running the test in isolation or with the entire
> > > suite?
> > > If with the entire suite, does the core dump occur during this test
> > > case or
> > > another?
> > >
> > > I've noticed that I occasionally get a "bus error" with the
> > > Big/JPEGEncodeInputTest.* tes

Re: [Libva] [Libva-intel-driver][PATCH] Use Media Read message if possible on Gen8+

2016-10-14 Thread Kelley, Sean V 


On 10/14/16, 2:12 PM, "Libva on behalf of Eoff, Ullysses A" 
<libva-boun...@lists.freedesktop.org on behalf of ullysses.a.e...@intel.com> 
wrote:

> -Original Message-
> From: Sean V Kelley [mailto:sea...@posteo.de]
> Sent: Friday, October 14, 2016 1:45 PM
> To: Eoff, Ullysses A <ullysses.a.e...@intel.com>; Zhao, Yakui 
<yakui.z...@intel.com>; Xiang, Haihao <haihao.xi...@intel.com>
> Cc: libva@lists.freedesktop.org
> Subject: Re: [Libva] [Libva-intel-driver][PATCH] Use Media Read message 
if possible on Gen8+
> 
> On Fri, 2016-10-14 at 19:58 +, Eoff, Ullysses A wrote:
> > > -Original Message-
> > > From: Libva [mailto:libva-boun...@lists.freedesktop.org] On Behalf
> > > Of Sean V Kelley
> > > Sent: Friday, October 14, 2016 11:44 AM
> > > To: Zhao, Yakui <yakui.z...@intel.com>; Xiang, Haihao  > > @intel.com>
> > > Cc: libva@lists.freedesktop.org
> > > Subject: Re: [Libva] [Libva-intel-driver][PATCH] Use Media Read
> > > message if possible on Gen8+
> > >
> > > On Fri, 2016-10-14 at 08:37 +0800, Zhao Yakui wrote:
> > > > On 10/13/2016 03:07 AM, Xiang, Haihao wrote:
> > > > > AVS can't gurantee bit-match for a large surface. This fixes
> > > > > the
> > > > > failure reported by gtest case
> > > > > Common/JPEGEncodeInputTest.Full/95.
> > > > >
> > > > > before:
> > > > > [  FAILED  ] Common/JPEGEncodeInputTest.Full/95, where
> > > > > GetParam() =
> > > > > (Fixed Size 7680x4320, 0x501176 pointing to "I420") (9239 ms)
> > > > > [--] 1 test from Common/JPEGEncodeInputTest (9239 ms
> > > > > total)
> > > > >
> > > > > [--] Global test environment tear-down
> > > > > [==] 1 test from 1 test case ran. (9361 ms total)
> > > > > [  PASSED  ] 0 tests.
> > > > > [  FAILED  ] 1 test, listed below:
> > > > > [  FAILED  ] Common/JPEGEncodeInputTest.Full/95, where
> > > > > GetParam() =
> > > > > (Fixed Size 7680x4320, 0x501176 pointing to "I420")
> > > > >
> > > > > after:
> > > > > [   OK ] Common/JPEGEncodeInputTest.Full/95 (15250 ms)
> > > > > [--] 1 test from Common/JPEGEncodeInputTest (15250 ms
> > > > > total)
> > > > >
> > > > > [--] Global test environment tear-down
> > > > > [==] 1 test from 1 test case ran. (15365 ms total)
> > > > > [  PASSED  ] 1 test.
> > > > >
> > > > > Signed-off-by: Xiang, Haihao<haihao.xi...@intel.com>
> > > >
> > > > This looks good to me.
> > > >
> > > > Add: Reviewed-by: Zhao Yakui <yakui.z...@intel.com>
> > >
> > >
> > > Getting a core dump although the test is now passing.  We'll need
> > > to
> > > debug further.
> > >
> >
> > I don't get any core dump on my SKL...  And the test passes for me
> > with
> > this patch.  Are you running the test in isolation or with the entire
> > suite?
> > If with the entire suite, does the core dump occur during this test
> > case or
> > another?
> >
> > I've noticed that I occasionally get a "bus error" with the
> > Big/JPEGEncodeInputTest.* test cases (i.e. 8192x8192) even without
> > this patch.  Perhaps this is what you're encountering?
> 
> I'm consistently seeing this on a SKL system.  It is also quite slow
> for execution.  Actually tested this on my Arch sytem with Kernel 4.7
> and the older LTS 4.4.  Same bus error and slow execution performance.
> 
> We need to isolate the root cause here because I suspect this is what
> Haihao has also been seeing.
> 
> Takes nearly a minute on one system and far less time on an Ubuntu
> system with the same kernel and essentially the same SKL SKU.
> 

Yes, I agree we need to isolate this issue.  I can file a bug for this but
was hoping to find time to generate a coredump output file.

The slow execution is caused by the byte comparison loop between the
input and the output YUV data... for la

Re: [Libva] [Libva-intel-driver][PATCH] Use Media Read message if possible on Gen8+

2016-10-14 Thread Eoff, Ullysses A 
> -Original Message-
> From: Sean V Kelley [mailto:sea...@posteo.de]
> Sent: Friday, October 14, 2016 1:45 PM
> To: Eoff, Ullysses A <ullysses.a.e...@intel.com>; Zhao, Yakui 
> <yakui.z...@intel.com>; Xiang, Haihao <haihao.xi...@intel.com>
> Cc: libva@lists.freedesktop.org
> Subject: Re: [Libva] [Libva-intel-driver][PATCH] Use Media Read message if 
> possible on Gen8+
> 
> On Fri, 2016-10-14 at 19:58 +, Eoff, Ullysses A wrote:
> > > -Original Message-
> > > From: Libva [mailto:libva-boun...@lists.freedesktop.org] On Behalf
> > > Of Sean V Kelley
> > > Sent: Friday, October 14, 2016 11:44 AM
> > > To: Zhao, Yakui <yakui.z...@intel.com>; Xiang, Haihao  > > @intel.com>
> > > Cc: libva@lists.freedesktop.org
> > > Subject: Re: [Libva] [Libva-intel-driver][PATCH] Use Media Read
> > > message if possible on Gen8+
> > >
> > > On Fri, 2016-10-14 at 08:37 +0800, Zhao Yakui wrote:
> > > > On 10/13/2016 03:07 AM, Xiang, Haihao wrote:
> > > > > AVS can't gurantee bit-match for a large surface. This fixes
> > > > > the
> > > > > failure reported by gtest case
> > > > > Common/JPEGEncodeInputTest.Full/95.
> > > > >
> > > > > before:
> > > > > [  FAILED  ] Common/JPEGEncodeInputTest.Full/95, where
> > > > > GetParam() =
> > > > > (Fixed Size 7680x4320, 0x501176 pointing to "I420") (9239 ms)
> > > > > [--] 1 test from Common/JPEGEncodeInputTest (9239 ms
> > > > > total)
> > > > >
> > > > > [--] Global test environment tear-down
> > > > > [==] 1 test from 1 test case ran. (9361 ms total)
> > > > > [  PASSED  ] 0 tests.
> > > > > [  FAILED  ] 1 test, listed below:
> > > > > [  FAILED  ] Common/JPEGEncodeInputTest.Full/95, where
> > > > > GetParam() =
> > > > > (Fixed Size 7680x4320, 0x501176 pointing to "I420")
> > > > >
> > > > > after:
> > > > > [   OK ] Common/JPEGEncodeInputTest.Full/95 (15250 ms)
> > > > > [--] 1 test from Common/JPEGEncodeInputTest (15250 ms
> > > > > total)
> > > > >
> > > > > [--] Global test environment tear-down
> > > > > [==] 1 test from 1 test case ran. (15365 ms total)
> > > > > [  PASSED  ] 1 test.
> > > > >
> > > > > Signed-off-by: Xiang, Haihao<haihao.xi...@intel.com>
> > > >
> > > > This looks good to me.
> > > >
> > > > Add: Reviewed-by: Zhao Yakui <yakui.z...@intel.com>
> > >
> > >
> > > Getting a core dump although the test is now passing.  We'll need
> > > to
> > > debug further.
> > >
> >
> > I don't get any core dump on my SKL...  And the test passes for me
> > with
> > this patch.  Are you running the test in isolation or with the entire
> > suite?
> > If with the entire suite, does the core dump occur during this test
> > case or
> > another?
> >
> > I've noticed that I occasionally get a "bus error" with the
> > Big/JPEGEncodeInputTest.* test cases (i.e. 8192x8192) even without
> > this patch.  Perhaps this is what you're encountering?
> 
> I'm consistently seeing this on a SKL system.  It is also quite slow
> for execution.  Actually tested this on my Arch sytem with Kernel 4.7
> and the older LTS 4.4.  Same bus error and slow execution performance.
> 
> We need to isolate the root cause here because I suspect this is what
> Haihao has also been seeing.
> 
> Takes nearly a minute on one system and far less time on an Ubuntu
> system with the same kernel and essentially the same SKL SKU.
> 

Yes, I agree we need to isolate this issue.  I can file a bug for this but
was hoping to find time to generate a coredump output file.

The slow execution is caused by the byte comparison loop between the
input and the output YUV data... for larger resolutions it takes a long time
and uses a lot of memory.  I don't like having long running tests, especially
at this level of testing...  so the only way to speed up this test is to 
optimize
the comparison routine.   Perhaps checksums would be better, but would
require a new library dependency.  We could also consider disabling this
test case (8192x8192) too.

Nonetheless, none of these issues are related to this patch since they
pre-existed.  Thus, this patch LGTM.

U. Artie

> Sean
> 
> 
> 
> >
&

Re: [Libva] [Libva-intel-driver][PATCH] Use Media Read message if possible on Gen8+

2016-10-14 Thread Sean V Kelley 
On Fri, 2016-10-14 at 19:58 +, Eoff, Ullysses A wrote:
> > -Original Message-
> > From: Libva [mailto:libva-boun...@lists.freedesktop.org] On Behalf
> > Of Sean V Kelley
> > Sent: Friday, October 14, 2016 11:44 AM
> > To: Zhao, Yakui <yakui.z...@intel.com>; Xiang, Haihao  > @intel.com>
> > Cc: libva@lists.freedesktop.org
> > Subject: Re: [Libva] [Libva-intel-driver][PATCH] Use Media Read
> > message if possible on Gen8+
> > 
> > On Fri, 2016-10-14 at 08:37 +0800, Zhao Yakui wrote:
> > > On 10/13/2016 03:07 AM, Xiang, Haihao wrote:
> > > > AVS can't gurantee bit-match for a large surface. This fixes
> > > > the
> > > > failure reported by gtest case
> > > > Common/JPEGEncodeInputTest.Full/95.
> > > > 
> > > > before:
> > > > [  FAILED  ] Common/JPEGEncodeInputTest.Full/95, where
> > > > GetParam() =
> > > > (Fixed Size 7680x4320, 0x501176 pointing to "I420") (9239 ms)
> > > > [--] 1 test from Common/JPEGEncodeInputTest (9239 ms
> > > > total)
> > > > 
> > > > [--] Global test environment tear-down
> > > > [==] 1 test from 1 test case ran. (9361 ms total)
> > > > [  PASSED  ] 0 tests.
> > > > [  FAILED  ] 1 test, listed below:
> > > > [  FAILED  ] Common/JPEGEncodeInputTest.Full/95, where
> > > > GetParam() =
> > > > (Fixed Size 7680x4320, 0x501176 pointing to "I420")
> > > > 
> > > > after:
> > > > [   OK ] Common/JPEGEncodeInputTest.Full/95 (15250 ms)
> > > > [--] 1 test from Common/JPEGEncodeInputTest (15250 ms
> > > > total)
> > > > 
> > > > [--] Global test environment tear-down
> > > > [==] 1 test from 1 test case ran. (15365 ms total)
> > > > [  PASSED  ] 1 test.
> > > > 
> > > > Signed-off-by: Xiang, Haihao<haihao.xi...@intel.com>
> > > 
> > > This looks good to me.
> > > 
> > > Add: Reviewed-by: Zhao Yakui <yakui.z...@intel.com>
> > 
> > 
> > Getting a core dump although the test is now passing.  We'll need
> > to
> > debug further.
> > 
> 
> I don't get any core dump on my SKL...  And the test passes for me
> with
> this patch.  Are you running the test in isolation or with the entire
> suite?
> If with the entire suite, does the core dump occur during this test
> case or
> another?
> 
> I've noticed that I occasionally get a "bus error" with the
> Big/JPEGEncodeInputTest.* test cases (i.e. 8192x8192) even without
> this patch.  Perhaps this is what you're encountering?

I'm consistently seeing this on a SKL system.  It is also quite slow
for execution.  Actually tested this on my Arch sytem with Kernel 4.7
and the older LTS 4.4.  Same bus error and slow execution performance.

We need to isolate the root cause here because I suspect this is what
Haihao has also been seeing.

Takes nearly a minute on one system and far less time on an Ubuntu
system with the same kernel and essentially the same SKL SKU.

Sean



> 
> U. Artie
> 
> > Sean
> > 
> > > 
> > > Thanks
> > >    Yakui
> > > 
> > > > ---
> > > >   src/gen8_post_processing.c | 56
> > > > +-
> > > >   src/shaders/post_processing/gen8/Makefile.am   |  2 +
> > > >   .../gen8/PL2_media_read_buf0123.g8a| 65
> > > > +
> > > >   .../gen8/PL3_media_read_buf0123.g8a| 68
> > > > ++
> > > >   src/shaders/post_processing/gen8/pl2_to_pl2.asm|  2 +
> > > >   src/shaders/post_processing/gen8/pl2_to_pl2.g8b| 44
> > > > ++
> > > >   src/shaders/post_processing/gen8/pl2_to_pl3.asm|  2 +
> > > >   src/shaders/post_processing/gen8/pl2_to_pl3.g8b| 44
> > > > ++
> > > >   src/shaders/post_processing/gen8/pl3_to_pl2.asm|  2 +
> > > >   src/shaders/post_processing/gen8/pl3_to_pl2.g8b| 47
> > > > +++
> > > >   src/shaders/post_processing/gen8/pl3_to_pl3.asm|  4 +-
> > > >   src/shaders/post_processing/gen8/pl3_to_pl3.g8b| 47
> > > > +++
> > > >   src/shaders/post_processing/gen9/pl2_to_pl2.g9b| 44
> > > > ++
> > > >   src/shaders/post_processing/gen

Re: [Libva] [Libva-intel-driver][PATCH] Use Media Read message if possible on Gen8+

2016-10-14 Thread Eoff, Ullysses A 

> -Original Message-
> From: Libva [mailto:libva-boun...@lists.freedesktop.org] On Behalf Of Sean V 
> Kelley
> Sent: Friday, October 14, 2016 11:44 AM
> To: Zhao, Yakui <yakui.z...@intel.com>; Xiang, Haihao <haihao.xi...@intel.com>
> Cc: libva@lists.freedesktop.org
> Subject: Re: [Libva] [Libva-intel-driver][PATCH] Use Media Read message if 
> possible on Gen8+
> 
> On Fri, 2016-10-14 at 08:37 +0800, Zhao Yakui wrote:
> > On 10/13/2016 03:07 AM, Xiang, Haihao wrote:
> > > AVS can't gurantee bit-match for a large surface. This fixes the
> > > failure reported by gtest case Common/JPEGEncodeInputTest.Full/95.
> > >
> > > before:
> > > [  FAILED  ] Common/JPEGEncodeInputTest.Full/95, where GetParam() =
> > > (Fixed Size 7680x4320, 0x501176 pointing to "I420") (9239 ms)
> > > [--] 1 test from Common/JPEGEncodeInputTest (9239 ms total)
> > >
> > > [--] Global test environment tear-down
> > > [==] 1 test from 1 test case ran. (9361 ms total)
> > > [  PASSED  ] 0 tests.
> > > [  FAILED  ] 1 test, listed below:
> > > [  FAILED  ] Common/JPEGEncodeInputTest.Full/95, where GetParam() =
> > > (Fixed Size 7680x4320, 0x501176 pointing to "I420")
> > >
> > > after:
> > > [   OK ] Common/JPEGEncodeInputTest.Full/95 (15250 ms)
> > > [--] 1 test from Common/JPEGEncodeInputTest (15250 ms
> > > total)
> > >
> > > [--] Global test environment tear-down
> > > [==] 1 test from 1 test case ran. (15365 ms total)
> > > [  PASSED  ] 1 test.
> > >
> > > Signed-off-by: Xiang, Haihao<haihao.xi...@intel.com>
> >
> > This looks good to me.
> >
> > Add: Reviewed-by: Zhao Yakui <yakui.z...@intel.com>
> 
> 
> Getting a core dump although the test is now passing.  We'll need to
> debug further.
> 

I don't get any core dump on my SKL...  And the test passes for me with
this patch.  Are you running the test in isolation or with the entire suite?
If with the entire suite, does the core dump occur during this test case or
another?

I've noticed that I occasionally get a "bus error" with the
Big/JPEGEncodeInputTest.* test cases (i.e. 8192x8192) even without
this patch.  Perhaps this is what you're encountering?

U. Artie

> Sean
> 
> >
> > Thanks
> >    Yakui
> >
> > > ---
> > >   src/gen8_post_processing.c | 56
> > > +-
> > >   src/shaders/post_processing/gen8/Makefile.am   |  2 +
> > >   .../gen8/PL2_media_read_buf0123.g8a| 65
> > > +
> > >   .../gen8/PL3_media_read_buf0123.g8a| 68
> > > ++
> > >   src/shaders/post_processing/gen8/pl2_to_pl2.asm|  2 +
> > >   src/shaders/post_processing/gen8/pl2_to_pl2.g8b| 44
> > > ++
> > >   src/shaders/post_processing/gen8/pl2_to_pl3.asm|  2 +
> > >   src/shaders/post_processing/gen8/pl2_to_pl3.g8b| 44
> > > ++
> > >   src/shaders/post_processing/gen8/pl3_to_pl2.asm|  2 +
> > >   src/shaders/post_processing/gen8/pl3_to_pl2.g8b| 47
> > > +++
> > >   src/shaders/post_processing/gen8/pl3_to_pl3.asm|  4 +-
> > >   src/shaders/post_processing/gen8/pl3_to_pl3.g8b| 47
> > > +++
> > >   src/shaders/post_processing/gen9/pl2_to_pl2.g9b| 44
> > > ++
> > >   src/shaders/post_processing/gen9/pl2_to_pl3.g9b| 44
> > > ++
> > >   src/shaders/post_processing/gen9/pl3_to_pl2.g9b| 47
> > > +++
> > >   src/shaders/post_processing/gen9/pl3_to_pl3.g9b| 47
> > > +++
> > >   16 files changed, 563 insertions(+), 2 deletions(-)
> > >   create mode 100644
> > > src/shaders/post_processing/gen8/PL2_media_read_buf0123.g8a
> > >   create mode 100644
> > > src/shaders/post_processing/gen8/PL3_media_read_buf0123.g8a
> > >
> > > diff --git a/src/gen8_post_processing.c
> > > b/src/gen8_post_processing.c
> > > index 375bbe0..687cedc 100644
> > > --- a/src/gen8_post_processing.c
> > > +++ b/src/gen8_post_processing.c
> > > @@ -630,6 +630,31 @@
> > > gen8_pp_set_media_rw_message_surface(VADriverContextP ctx, struct
> > > i965_post_proc
> > >  SURFACE_FORMAT_R8_UNORM,
> &

Re: [Libva] [Libva-intel-driver][PATCH] Use Media Read message if possible on Gen8+

2016-10-14 Thread Sean V Kelley 
On Fri, 2016-10-14 at 08:37 +0800, Zhao Yakui wrote:
> On 10/13/2016 03:07 AM, Xiang, Haihao wrote:
> > AVS can't gurantee bit-match for a large surface. This fixes the
> > failure reported by gtest case Common/JPEGEncodeInputTest.Full/95.
> > 
> > before:
> > [  FAILED  ] Common/JPEGEncodeInputTest.Full/95, where GetParam() =
> > (Fixed Size 7680x4320, 0x501176 pointing to "I420") (9239 ms)
> > [--] 1 test from Common/JPEGEncodeInputTest (9239 ms total)
> > 
> > [--] Global test environment tear-down
> > [==] 1 test from 1 test case ran. (9361 ms total)
> > [  PASSED  ] 0 tests.
> > [  FAILED  ] 1 test, listed below:
> > [  FAILED  ] Common/JPEGEncodeInputTest.Full/95, where GetParam() =
> > (Fixed Size 7680x4320, 0x501176 pointing to "I420")
> > 
> > after:
> > [   OK ] Common/JPEGEncodeInputTest.Full/95 (15250 ms)
> > [--] 1 test from Common/JPEGEncodeInputTest (15250 ms
> > total)
> > 
> > [--] Global test environment tear-down
> > [==] 1 test from 1 test case ran. (15365 ms total)
> > [  PASSED  ] 1 test.
> > 
> > Signed-off-by: Xiang, Haihao
> 
> This looks good to me.
> 
> Add: Reviewed-by: Zhao Yakui 


Getting a core dump although the test is now passing.  We'll need to
debug further.

Sean

> 
> Thanks
>    Yakui
> 
> > ---
> >   src/gen8_post_processing.c | 56
> > +-
> >   src/shaders/post_processing/gen8/Makefile.am   |  2 +
> >   .../gen8/PL2_media_read_buf0123.g8a| 65
> > +
> >   .../gen8/PL3_media_read_buf0123.g8a| 68
> > ++
> >   src/shaders/post_processing/gen8/pl2_to_pl2.asm|  2 +
> >   src/shaders/post_processing/gen8/pl2_to_pl2.g8b| 44
> > ++
> >   src/shaders/post_processing/gen8/pl2_to_pl3.asm|  2 +
> >   src/shaders/post_processing/gen8/pl2_to_pl3.g8b| 44
> > ++
> >   src/shaders/post_processing/gen8/pl3_to_pl2.asm|  2 +
> >   src/shaders/post_processing/gen8/pl3_to_pl2.g8b| 47
> > +++
> >   src/shaders/post_processing/gen8/pl3_to_pl3.asm|  4 +-
> >   src/shaders/post_processing/gen8/pl3_to_pl3.g8b| 47
> > +++
> >   src/shaders/post_processing/gen9/pl2_to_pl2.g9b| 44
> > ++
> >   src/shaders/post_processing/gen9/pl2_to_pl3.g9b| 44
> > ++
> >   src/shaders/post_processing/gen9/pl3_to_pl2.g9b| 47
> > +++
> >   src/shaders/post_processing/gen9/pl3_to_pl3.g9b| 47
> > +++
> >   16 files changed, 563 insertions(+), 2 deletions(-)
> >   create mode 100644
> > src/shaders/post_processing/gen8/PL2_media_read_buf0123.g8a
> >   create mode 100644
> > src/shaders/post_processing/gen8/PL3_media_read_buf0123.g8a
> > 
> > diff --git a/src/gen8_post_processing.c
> > b/src/gen8_post_processing.c
> > index 375bbe0..687cedc 100644
> > --- a/src/gen8_post_processing.c
> > +++ b/src/gen8_post_processing.c
> > @@ -630,6 +630,31 @@
> > gen8_pp_set_media_rw_message_surface(VADriverContextP ctx, struct
> > i965_post_proc
> >  SURFACE_FORMAT_R8_UNORM,
> > 0,
> >  base_index + 2);
> >   }
> > +
> > +gen8_pp_set_surface_state(ctx, pp_context,
> > +  bo, 0,
> > +  ALIGN(width[0], 4) / 4,
> > height[0], pitch[0],
> > +  I965_SURFACEFORMAT_R8_UINT,
> > +  base_index + 3, 1);
> > +
> > +if (fourcc_info->num_planes == 2) {
> > +gen8_pp_set_surface_state(ctx, pp_context,
> > +  bo, offset[1],
> > +  ALIGN(width[1], 2) / 2,
> > height[1], pitch[1],
> > +  I965_SURFACEFORMAT_R8G8_SINT
> > ,
> > +  base_index + 4, 1);
> > +} else if (fourcc_info->num_planes == 3) {
> > +gen8_pp_set_surface_state(ctx, pp_context,
> > +  bo, offset[1],
> > +  ALIGN(width[1], 4) / 4,
> > height[1], pitch[1],
> > +  I965_SURFACEFORMAT_R8_SINT,
> > +  base_index + 4, 1);
> > +gen8_pp_set_surface_state(ctx, pp_context,
> > +  bo, offset[2],
> > +  ALIGN(width[2], 4) / 4,
> > height[2], pitch[2],
> > +  I965_SURFACEFORMAT_R8_SINT,
> > +  base_index + 5, 1);
> > +}
> >   }
> >   }
> > 
> > @@ -788,6 +813,33 @@ gen8_pp_get_8tap_filter_mode(VADriverContextP
> > ctx,
> >   return 3;
> >   }
> > 
> > +static int
> > +gen8_pp_kernel_use_media_read_msg(VADriverContextP ctx,
> > +  

Re: [Libva] [Libva-intel-driver][PATCH] Use Media Read message if possible on Gen8+

2016-10-13 Thread Zhao Yakui 

On 10/13/2016 03:07 AM, Xiang, Haihao wrote:

AVS can't gurantee bit-match for a large surface. This fixes the
failure reported by gtest case Common/JPEGEncodeInputTest.Full/95.

before:
[  FAILED  ] Common/JPEGEncodeInputTest.Full/95, where GetParam() = (Fixed Size 
7680x4320, 0x501176 pointing to "I420") (9239 ms)
[--] 1 test from Common/JPEGEncodeInputTest (9239 ms total)

[--] Global test environment tear-down
[==] 1 test from 1 test case ran. (9361 ms total)
[  PASSED  ] 0 tests.
[  FAILED  ] 1 test, listed below:
[  FAILED  ] Common/JPEGEncodeInputTest.Full/95, where GetParam() = (Fixed Size 
7680x4320, 0x501176 pointing to "I420")

after:
[   OK ] Common/JPEGEncodeInputTest.Full/95 (15250 ms)
[--] 1 test from Common/JPEGEncodeInputTest (15250 ms total)

[--] Global test environment tear-down
[==] 1 test from 1 test case ran. (15365 ms total)
[  PASSED  ] 1 test.

Signed-off-by: Xiang, Haihao


This looks good to me.

Add: Reviewed-by: Zhao Yakui 

Thanks
  Yakui


---
  src/gen8_post_processing.c | 56 +-
  src/shaders/post_processing/gen8/Makefile.am   |  2 +
  .../gen8/PL2_media_read_buf0123.g8a| 65 +
  .../gen8/PL3_media_read_buf0123.g8a| 68 ++
  src/shaders/post_processing/gen8/pl2_to_pl2.asm|  2 +
  src/shaders/post_processing/gen8/pl2_to_pl2.g8b| 44 ++
  src/shaders/post_processing/gen8/pl2_to_pl3.asm|  2 +
  src/shaders/post_processing/gen8/pl2_to_pl3.g8b| 44 ++
  src/shaders/post_processing/gen8/pl3_to_pl2.asm|  2 +
  src/shaders/post_processing/gen8/pl3_to_pl2.g8b| 47 +++
  src/shaders/post_processing/gen8/pl3_to_pl3.asm|  4 +-
  src/shaders/post_processing/gen8/pl3_to_pl3.g8b| 47 +++
  src/shaders/post_processing/gen9/pl2_to_pl2.g9b| 44 ++
  src/shaders/post_processing/gen9/pl2_to_pl3.g9b| 44 ++
  src/shaders/post_processing/gen9/pl3_to_pl2.g9b| 47 +++
  src/shaders/post_processing/gen9/pl3_to_pl3.g9b| 47 +++
  16 files changed, 563 insertions(+), 2 deletions(-)
  create mode 100644 src/shaders/post_processing/gen8/PL2_media_read_buf0123.g8a
  create mode 100644 src/shaders/post_processing/gen8/PL3_media_read_buf0123.g8a

diff --git a/src/gen8_post_processing.c b/src/gen8_post_processing.c
index 375bbe0..687cedc 100644
--- a/src/gen8_post_processing.c
+++ b/src/gen8_post_processing.c
@@ -630,6 +630,31 @@ gen8_pp_set_media_rw_message_surface(VADriverContextP ctx, 
struct i965_post_proc
 SURFACE_FORMAT_R8_UNORM, 0,
 base_index + 2);
  }
+
+gen8_pp_set_surface_state(ctx, pp_context,
+  bo, 0,
+  ALIGN(width[0], 4) / 4, height[0], pitch[0],
+  I965_SURFACEFORMAT_R8_UINT,
+  base_index + 3, 1);
+
+if (fourcc_info->num_planes == 2) {
+gen8_pp_set_surface_state(ctx, pp_context,
+  bo, offset[1],
+  ALIGN(width[1], 2) / 2, height[1], 
pitch[1],
+  I965_SURFACEFORMAT_R8G8_SINT,
+  base_index + 4, 1);
+} else if (fourcc_info->num_planes == 3) {
+gen8_pp_set_surface_state(ctx, pp_context,
+  bo, offset[1],
+  ALIGN(width[1], 4) / 4, height[1], 
pitch[1],
+  I965_SURFACEFORMAT_R8_SINT,
+  base_index + 4, 1);
+gen8_pp_set_surface_state(ctx, pp_context,
+  bo, offset[2],
+  ALIGN(width[2], 4) / 4, height[2], 
pitch[2],
+  I965_SURFACEFORMAT_R8_SINT,
+  base_index + 5, 1);
+}
  }
  }

@@ -788,6 +813,33 @@ gen8_pp_get_8tap_filter_mode(VADriverContextP ctx,
  return 3;
  }

+static int
+gen8_pp_kernel_use_media_read_msg(VADriverContextP ctx,
+  const struct i965_surface *src_surface,
+  const VARectangle *src_rect,
+  const struct i965_surface *dst_surface,
+  const VARectangle *dst_rect)
+{
+int src_fourcc = pp_get_surface_fourcc(ctx, src_surface);
+int dst_fourcc = pp_get_surface_fourcc(ctx, dst_surface);
+const i965_fourcc_info *src_fourcc_info = get_fourcc_info(src_fourcc);
+const i965_fourcc_info *dst_fourcc_info = get_fourcc_info(dst_fourcc);
+
+if (!src_fourcc_info ||
+

[Libva] [Libva-intel-driver][PATCH] Use Media Read message if possible on Gen8+

2016-10-12 Thread Xiang, Haihao 
AVS can't gurantee bit-match for a large surface. This fixes the
failure reported by gtest case Common/JPEGEncodeInputTest.Full/95.

before:
[  FAILED  ] Common/JPEGEncodeInputTest.Full/95, where GetParam() = (Fixed Size 
7680x4320, 0x501176 pointing to "I420") (9239 ms)
[--] 1 test from Common/JPEGEncodeInputTest (9239 ms total)

[--] Global test environment tear-down
[==] 1 test from 1 test case ran. (9361 ms total)
[  PASSED  ] 0 tests.
[  FAILED  ] 1 test, listed below:
[  FAILED  ] Common/JPEGEncodeInputTest.Full/95, where GetParam() = (Fixed Size 
7680x4320, 0x501176 pointing to "I420")

after:
[   OK ] Common/JPEGEncodeInputTest.Full/95 (15250 ms)
[--] 1 test from Common/JPEGEncodeInputTest (15250 ms total)

[--] Global test environment tear-down
[==] 1 test from 1 test case ran. (15365 ms total)
[  PASSED  ] 1 test.

Signed-off-by: Xiang, Haihao 
---
 src/gen8_post_processing.c | 56 +-
 src/shaders/post_processing/gen8/Makefile.am   |  2 +
 .../gen8/PL2_media_read_buf0123.g8a| 65 +
 .../gen8/PL3_media_read_buf0123.g8a| 68 ++
 src/shaders/post_processing/gen8/pl2_to_pl2.asm|  2 +
 src/shaders/post_processing/gen8/pl2_to_pl2.g8b| 44 ++
 src/shaders/post_processing/gen8/pl2_to_pl3.asm|  2 +
 src/shaders/post_processing/gen8/pl2_to_pl3.g8b| 44 ++
 src/shaders/post_processing/gen8/pl3_to_pl2.asm|  2 +
 src/shaders/post_processing/gen8/pl3_to_pl2.g8b| 47 +++
 src/shaders/post_processing/gen8/pl3_to_pl3.asm|  4 +-
 src/shaders/post_processing/gen8/pl3_to_pl3.g8b| 47 +++
 src/shaders/post_processing/gen9/pl2_to_pl2.g9b| 44 ++
 src/shaders/post_processing/gen9/pl2_to_pl3.g9b| 44 ++
 src/shaders/post_processing/gen9/pl3_to_pl2.g9b| 47 +++
 src/shaders/post_processing/gen9/pl3_to_pl3.g9b| 47 +++
 16 files changed, 563 insertions(+), 2 deletions(-)
 create mode 100644 src/shaders/post_processing/gen8/PL2_media_read_buf0123.g8a
 create mode 100644 src/shaders/post_processing/gen8/PL3_media_read_buf0123.g8a

diff --git a/src/gen8_post_processing.c b/src/gen8_post_processing.c
index 375bbe0..687cedc 100644
--- a/src/gen8_post_processing.c
+++ b/src/gen8_post_processing.c
@@ -630,6 +630,31 @@ gen8_pp_set_media_rw_message_surface(VADriverContextP ctx, 
struct i965_post_proc
SURFACE_FORMAT_R8_UNORM, 0,
base_index + 2);
 }
+
+gen8_pp_set_surface_state(ctx, pp_context,
+  bo, 0,
+  ALIGN(width[0], 4) / 4, height[0], pitch[0],
+  I965_SURFACEFORMAT_R8_UINT,
+  base_index + 3, 1);
+
+if (fourcc_info->num_planes == 2) {
+gen8_pp_set_surface_state(ctx, pp_context,
+  bo, offset[1],
+  ALIGN(width[1], 2) / 2, height[1], 
pitch[1],
+  I965_SURFACEFORMAT_R8G8_SINT,
+  base_index + 4, 1);
+} else if (fourcc_info->num_planes == 3) {
+gen8_pp_set_surface_state(ctx, pp_context,
+  bo, offset[1],
+  ALIGN(width[1], 4) / 4, height[1], 
pitch[1],
+  I965_SURFACEFORMAT_R8_SINT,
+  base_index + 4, 1);
+gen8_pp_set_surface_state(ctx, pp_context,
+  bo, offset[2],
+  ALIGN(width[2], 4) / 4, height[2], 
pitch[2],
+  I965_SURFACEFORMAT_R8_SINT,
+  base_index + 5, 1);
+}
 }
 }
 
@@ -788,6 +813,33 @@ gen8_pp_get_8tap_filter_mode(VADriverContextP ctx,
 return 3;
 }
 
+static int
+gen8_pp_kernel_use_media_read_msg(VADriverContextP ctx,
+  const struct i965_surface *src_surface,
+  const VARectangle *src_rect,
+  const struct i965_surface *dst_surface,
+  const VARectangle *dst_rect)
+{
+int src_fourcc = pp_get_surface_fourcc(ctx, src_surface);
+int dst_fourcc = pp_get_surface_fourcc(ctx, dst_surface);
+const i965_fourcc_info *src_fourcc_info = get_fourcc_info(src_fourcc);
+const i965_fourcc_info *dst_fourcc_info = get_fourcc_info(dst_fourcc);
+
+if (!src_fourcc_info ||
+src_fourcc_info->subsampling != SUBSAMPLE_YUV420 ||
+!dst_fourcc_info ||
+dst_fourcc_info->subsampling != SUBSAMPLE_YUV420)
+return 0;
+
+if