hardware-intel-libva

Fork

va_trace: allow surface trace to dump part of surface content

To debug/isolate video record quality issue (blocky/corruption, etc),
we always need to dump the YUV image from camera. LibVA has such
functionality, but due to storage bandwidth issue (~10M Byte/second),
it is hard to write 1080P 30fps YUV data to disk (~88M Byte/second).
Now refine the functionality to only dump part of the image, and
won't need to modify the code in the default build.

The method is:
edit /etc/libva.conf to include bellow (all export them):
LIBVA_TRACE_SURFACE=/data/enc-XXX.yuv
LIBVA_TRACE_GEOMETRY=320x240+20+20

It will only save 320x240 from (20,20) of the source surface to file
/data/enc-XXX.yuv.

Signed-off-by: Austin Yuan <shengquan.yuan@intel.com>
(cherry picked from commit 5190a7465756a41808a562c8eccec521dbf77d8c)

va_trace_surface: dumpe surface content after vaEndPicture call

vaEndPicture will kick HW decode/encode, and at this time use CPU
to dump surface content can keep GPU/CPU in parallel.

Also for encode trace, don't need to wait current frame completions
but for decode, we must wait decode till it is completed

Signed-off-by: Austin Yuan <shengquan.yuan@intel.com>
(cherry picked from commit 0af365de4c03c12fc20183093a944f7513a1c7be)

Fix va_fool.c klocwork issue

BZ: 83882

Fix klocwork issue that complaine array 'fool_context[idx].fool_buf'
of size 43 may use index value(s) 0..255

Change-Id: I196a36c0067e2b7e390567a203ef1520af5b873f
Signed-off-by: Elaine Wang <elaine.wang@intel.com>

klockwork: Fix libva klocwork scaning issues

Fix NULL pointer dereference risk issues and uninitialized variables

Signed-off-by: Elaine Wang <elaine.wang@intel.com>

Change-Id: I839a9def7b08c41c1f075e4cd1cc06e8bb0c8c9e

tests: cope with new vaQuerySurfaceAttributes() API.

Also factor the code to check whether a specific image format or
surface attribute exists. Cache the list of supported image formats
and surface attributes.

Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>

test: add color conversion test option in putsurface

- command line: ./putsurface --fmt1 format1 --fmt2 format2
or ./putsurface -1 format1 -2 format2
- example: ./putsurface -1 NV12 -2 YV12
- implementation:
- vaSurface is created with fmt1
- get image (with fmt2) from vaSurface,
it leads to color conversion
- put the above image to a temp surface
color conversion may happen
- render the above temp surface instead of original one

tests: cope with new vaCreateSurfaces() API.

Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
Signed-off-by: Haitao Huang <haitao.huang@intel.com>
Signed-off-by: Austin Yuan <shengquan.yuan@intel.com>

Add default implementation for vaQuerySurfaceAttributes().

Add an initial implementation for vaQuerySurfaceAttributes() in terms
of vaGetSurfaceAttributes(). This is an inefficient approach because
it dynamically allocates the whole set of surface attributes we guessed
are available.

This code is to be removed once the VA drivers are updated.

Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>

API: add vaQuerySurfaceAttributes().

vaQuerySurfaceAttributes() is the recommended way to query the driver for
all surface attributes that are supported by the underlying hardware. This
is more useful than guessing the attribute values and testing them through
vaGetSurfaceAttributes().

The older vaGetSurfaceAttributes() function is hereby removed. However, the
relevant VA driver hook is kept for now, and will be removed/replaced for a
future update.

Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>

Added new VA_RT_FORMAT_RGBP and VA_FOURCC_411R

SurfaceAttribMemoryType: add ION memory type for Android

ION is the new Android memory manager for cross-components buffer
sharing. Adding this memory type for it.

Signed-off-by: Austin Yuan <shengquan.yuan@intel.com>

ASurfaceAttribExternalBuffers flags: add more memory attribute flags

The external user pointer buffers may be allocated as cacheable or
non-cacheable, with the addtitional cache attribute flags, it can help
driver implementation to set correct cache attribute into device
memory controller (e.g. MMU)

Signed-off-by: Austin Yuan <shengquan.yuan@intel.com>

Changed values for VASurfaceAttribMemoryType from fourcc to bit fields.

Added new RT formats and fourcc codes for pixel formats.

Added new surface attributes to enable VA surface creation from external buffers.

Add RGB RT format definition for video process

RGB format surface is required for color
space conversion from YUV to RGB.

signed-off-by: Li,Xiaowei <xiaowei.a.li@intel.com>

tpi: clean the old interface, and use new one

Signed-off-by: Austin Yuan <shengquan.yuan@intel.com>

Add new fourcc codes

Added fourcc codes for YV24 and YV32 (4:4:4 planar formats)

Signed-off-by: Jonathan Bian <jonathan.bian@intel.com>

va_TraceCreateSurface->va_TraceCreateSurfaces: correct name change

Signed-off-by: Austin Yuan <shengquan.yuan@intel.com>

API: add compat glue for older vaCreateSurfaces().

API: add surface attributes.

test: update MPEG-2 encoding case

Signed-off-by: Xiang, Haihao <haihao.xiang@intel.com>

tests: disable v4l demo for now.

vaPutSurface test on Android: fix build isste on Android JB

Change-Id: Ibfb3d6a653f9af924422edb052e751b84015ffd0
Signed-off-by: Austin Yuan <shengquan.yuan@gmail.com>

h264encode: fix build warning on Android

Change-Id: Iee2886bde2362e75d7e1837211553d9672bba650
Signed-off-by: Austin Yuan <shengquan.yuan@gmail.com>

h264encode: remove the leftover file h264encode_android.cpp

Signed-off-by: Austin Yuan <shengquan.yuan@gmail.com>

h264encode: fix some issues

1. reference frame setting looks odd
2. fix the issue that the first 16 frames are encoded twice

Signed-off-by: Elaine Wang <elaine.wang@intel.com>

test/h264encode: refine the h264encode to support advanced mode

Signed-off-by: Austin Yuan <shengquan.yuan@gmail.com>

test/putsurface_x11.c: don't hard code the DISPLAY to ":0.0"

Signed-off-by: Austin Yuan <shengquan.yuan@gmail.com>

mpeg2enc: find the proper profile and level

Signed-off-by: Xiang, Haihao <haihao.xiang@intel.com>