Age | Commit message (Collapse) | Author | Files | Lines |
|
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
|
|
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
|
|
Most OSVs adopted XXX-intel-driver as their shipping packages for the
Intel VA driver, with XXX = { vaapi, libva }. Adopt libva-intel-driver
from now on for upstream packages too.
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
|
|
Signed-off-by: Xiang, Haihao <haihao.xiang@intel.com>
|
|
Signed-off-by: Xiang, Haihao <haihao.xiang@intel.com>
|
|
Currently the HW only supprts JPEG baseline decoding
Signed-off-by: Xiang, Haihao <haihao.xiang@intel.com>
(cherry picked from commit d4be016d054ba79f5b4882088ded96528a60c985)
Conflicts:
src/gen7_mfd.c
|
|
This reverts commit bcca7d23de29423b0d7d9954bfb21cc9bf3834f5.
|
|
Sync changes from 1.0.x series.
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
|
|
Preparatory work for some changes to be promoted from staging to master.
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
|
|
By default, newer automake versions make tar use the ancient tar V7 format
that limits filenames to 99 characters. This is troublesome for certain shader
sources files. Now require automake 1.9 and use the tar-ustar option so that
a newer format, defined in POSIX 1003.1-1988, is used. The limitation is now
256 characters.
PAX interchange format (POSIX 1003.1-2001) was considered but probably still
to young? Just stick to the intermediate format for now as this is enough for
our purposes.
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
|
|
Fix decoding of 4K videos on Ivy Bridge. Note that either dimension
shall not exceed 4096 pixels.
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
|
|
|
|
Signed-off-by: Xiang, Haihao <haihao.xiang@intel.com>
|
|
Signed-off-by: Xiang, Haihao <haihao.xiang@intel.com>
|
|
Thank Gwenole for pointing it out
Signed-off-by: Xiang, Haihao <haihao.xiang@intel.com>
|
|
If load_intra_quantiser_matrix/load_non_intra_quantiser_matrix is set to 0, then
there is no change to quantisation matrix. In this case, submit MFX_QM_STATE
with previous setting
Signed-off-by: Xiang, Haihao <haihao.xiang@intel.com>
|
|
Signed-off-by: Xiang, Haihao <haihao.xiang@intel.com>
|
|
Signed-off-by: Xiang, Haihao <haihao.xiang@intel.com>
|
|
It won't build JPEG without config.h
Signed-off-by: Xiang, Haihao <haihao.xiang@intel.com>
|
|
Fill in remaining reference surfaces as recommanded in HW specs.
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
|
|
It will avoid GPU hang when try to play unsupported large resolution
videos.
Signed-off-by: Xiang, Haihao <haihao.xiang@intel.com>
|
|
The size is scalable with frame width or height
Signed-off-by: Xiang, Haihao <haihao.xiang@intel.com>
|
|
Signed-off-by: Xiang, Haihao <haihao.xiang@intel.com>
|
|
Fix construction of reference frames list for interlaced streams.
In this case, the array is indexed by frame store ID >> 1 where
bit 0 of frame store ID represents top (0) or bottom (1) field.
Besides, current render target can also be used as a reference
while decoding the second field.
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
|
|
Avoid an assert() since we were assigning a reference surface even
if it did not have any backing store.
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
|
|
Gen6 has specific requirements for the TFF flag, and thus has different
semantics than Gen7 (IVB). In particular, HW uses picture_structure and
TFF flag to determine the correct field to render.
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
|
|
Correctly emit slice_vertical_position, as per the definition from
the bitstream, to the HW decoder (MFD_MPEG2_BSD_OBJECT).
Add workaround for players that have not fixed their usage of
slice_vertical_position. That field shall represent the slice vertical
position as it comes from the bitstream.
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
|
|
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
|
|
Handle bob-deinterlacing flags passed to vaPutSurface().
i.e. VA_TOP_FIELD|VA_BOTTOM_FIELD.
Avoid advanced deinterlacing kernels as they allocate extra temporary
surfaces, which are useless for such simple tasks. i.e. display either
field of an interlaced surface.
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
|
|
... aka fix wrong patch applied. For the records, considering header
size without EPB is N, then the max header size with EPB is N * 3/2.
However, it's not possible to reach this value since that would mean
that slice_header() only contained zeroes.
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
|
|
Handle two indices in parallel, one relative to the bitstream with
EPB, and the other without EPB.
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
|
|
If weighted_bipred_idc == 2, luma and chroma weight denom (in log2 base)
shall be set to the default value on Sandy Bridge. i.e. 5 as per defined
at 8-279 in AVC specs.
https://bugs.freedesktop.org/show_bug.cgi?id=40820
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
|
|
Implicit weight tables derivation (weighted_bipred_idc == 2) could
not be set. Only default weight tables were possible instead of
explicit weight tables provided by the codec layer.
https://bugs.freedesktop.org/show_bug.cgi?id=40820
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
|
|
Fix and simplify the scan for emulation_prevention_bytes, thus avoiding
a read beyond the end of the slice data buffer. Besides, if slice_header()
bytes are needed, use dri_bo_get_subdata() instead.
HW specific changes:
- SNB: make the HW skip the emulation prevention bytes itself.
- IVB: fix MFD_AVC_BSD_OBJECT to report the actual slice data buffer size.
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
|
|
This fixes buffer overflow in the newly allocated Gen buffer that holds
VC-1 bitplanes.
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
|
|
If codec layer does not provide a VAIQMatrixBufferH264, this means
flat scaling lists shall be used. The MFX_QM_STATE command still has
to be submitted since IVB+ does not have a means to use HW generated
scaling lists.
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
|
|
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
|
|
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
|
|
Signed-off-by: Xiang, Haihao <haihao.xiang@intel.com>
|
|
Signed-off-by: Xiang, Haihao <haihao.xiang@intel.com>
|
|
Signed-off-by: Xiang, Haihao <haihao.xiang@intel.com>
|
|
Signed-off-by: Xiang, Haihao <haihao.xiang@intel.com>
|
|
Signed-off-by: Xiang, Haihao <haihao.xiang@intel.com>
|
|
Signed-off-by: Xiang, Haihao <haihao.xiang@intel.com>
|
|
Signed-off-by: Xiang, Haihao <haihao.xiang@intel.com>
|
|
Signed-off-by: Xiang, Haihao <haihao.xiang@intel.com>
|
|
Signed-off-by: Xiang, Haihao <haihao.xiang@intel.com>
|
|
When call i965_BeginPicture for JPEG decoding, Assertion `0' failed.
Signed-off-by: Hai Lan <hai.lan@intel.com>
|
|
Signed-off-by: Xiang, Haihao <haihao.xiang@intel.com>
|
|
Signed-off-by: Xiang, Haihao <haihao.xiang@intel.com>
|