Age | Commit message (Collapse) | Author |
|
bug 2247143
|
|
bug 2247143
|
|
* changes:
Jpeg encoder: remove pplib support
|
|
bug 2100370
Originally from: https://partner.source.android.com/g/#change,1410
rebased...
|
|
* changes:
Fix in h264 output buffer size calculation
|
|
This removes support for rgb input and a 20MB allocation. The
result is less memory usage during encoding a jpeg image and
overall improved system performance.
GID1456. b2237074
|
|
Previous buffer calculation was incorrect, for D1 it was 2.1 MB now
it is 600 KB. The calculations are now based on H.264 annex A spec.
The smaller size results in reducing cache flush latencies in the bridge improving
performance. With this change D1 can now get up to ~29 FPS.
Credit to Jeff Vanhoof for noticing the large buffer size.
Originally from: https://partner.source.android.com/g/#change,1400
|
|
2. Modified the original patch to add error handling for DSP Initialization failure
to prevent OMX client hang
Originally from: https://partner.source.android.com/g/#change,1411
|
|
Dynamic framerate exposed a bug in OMX as the algorithm started producing
frames with multiple NAL units. OMX was tagging correct time stamp only
to last NAL unit in each frame. This has been fixed now to tag the same
timestamp to all NAL units belonging to each frame. Looks like PV muxer
and demuxer are compatible and apparently this is not an issue for them.
But quicktime and VLC don't like NAL units with 0 timestamp in the
middle of a bitstream.
Originally from: https://partner.source.android.com/g/#change,1402
|
|
Couple of input params were missing on the OMX side.
The SN accesses an unmapped region and faults Added
ulQpMax,ulQpMin to MP4VE_GPP_SN_UALGInputParams reduced
the size of ResyncData to 1620 to match SN header file
Originally from: https://partner.source.android.com/g/#change,1401
|
|
Return an error event in the case when the encoder components are unable to
transition to the idle state.
Originally from: https://partner.source.android.com/g/#change,1403
|
|
bug 2174390
1. This will protect buffers from getting unmapped from one thread while other thread is mapping them.
2. It will also protect strmctrl & algctrl related structures from getting unmapped while other thread is using them for sending message
to DSP.
This is a partial solution for testVideoCapture failures. Refer to CSR# OMAPS00207906.
Incorporates fixes for review comments made by Dave Sparks.
This patch replaces the original patch #1287.
Signed-off-by: Sanjay Guin <sguin1@ti.com>
Signed-off-by: Daniel Charles <dcharles@ti.com>
|
|
bug 2168700
Changes include -
1. For higher resolution encoding preset set to 4 for optimized performance.
It enables following optimizations on DSP side -
- Modified HW Accelerator scheduling on sequencer,
- Hierarchical P frame encoding,
- Modified skip MB decision,
- Modified Intra mode selection scheme
- Instruction cache modifications.
2. Constant bit rate control set for high resolutions.
3. maxMBsPerSlice & maxBytesPerSlice set to 0
4. Deblocking disabled at Sn create time for nresolutions higher than VGA.
Signed-off-by: Nikhil Mande <nikhil@ti.com>
Signed-off-by: Jorge Solano <jsolano@ti.com>
Originally from: https://partner.source.android.com/g/#change,1142
Did manual merge.
|
|
encoder
Originally from: https://partner.source.android.com/g/#change,1218
|
|
encoder
Originally from: https://partner.source.android.com/g/#change,1214
|
|
encoder
Originally from: https://partner.source.android.com/g/#change,1211
|
|
encoder
Originally from: https://partner.source.android.com/g/#change,1210
|
|
decode
Originally from: https://partner.source.android.com/g/#change,1201
|
|
encode
Originally from: https://partner.source.android.com/g/#change,1200
|
|
Originally from: https://partner.source.android.com/g/#change,1150
|
|
Originally from: https://partner.source.android.com/g/#change,1149
|
|
Originally from: https://partner.source.android.com/g/#change,1135
|
|
Fixed a typo in the patch 27455
|
|
This reverts commit 9961abce63e3f897758a99cc2e060cb68c484803.
|
|
Originally from: https://partner.source.android.com/g/#change,1177
Fixed assert that has missed in the original patch.
|
|
* changes:
Fix OMX_GetRolesOfComponents and OMX_GetComponentsOfRole, and clean up some unused code.
|
|
* changes:
Don't build the TI OMX test binaries by default In addition, do not build the audio decoders, jpeg decoder, and VPP.
|
|
Changes made to LCML:
1. Modified append_dsp_path to allocate one more byte for the null terminator. (ROOT CAUSE)
2. Added missing memset for certain malloc calls.
3. Added macros for malloc and free - which would check for NULL pointer.
4. Ensure the create phase array copy does not continue to look for the MARKER past the size of the array.
Signed-off-by: Anu Sundararajan <sanuradha@ti.com>
|
|
In addition, do not build the audio decoders, jpeg decoder, and VPP.
|
|
unused code.
Signed-off-by: Daniel Charles <dcharles@ti.com>
Orignally from: https://partner.source.android.com/g/#change,1171
Modified based on review feedbacks and comments
|
|
|
|
Limit simultaneous instances per component
-Defined MAX_CONCURRENT_INSTANCES=1 in ComponentRegistry.h.
-This macro can be increased to allow more instances per codec
-JPEG.Encoder library typo on component list.
- Lookingfor the correct amount of components on the component list to avoid a NULL pointer comparation.
Originally from: https://partner.source.android.com/g/#change,1132
Also take some comments and feedbacks from reviewers
|
|
This reverts commit a5a79c7a1090c1c02aa71c847a17dced8e119ddb.
|
|
* changes:
Thread safety of reference counter OMX Compliancy issues Limit simultaneous instances per component -Defined MAX_CONCURRENT_INSTANCES=1 in ComponentRegistry.h -this macro can be increased to allow more instances per codec
|
|
OMX Compliancy issues
Limit simultaneous instances per component
-Defined MAX_CONCURRENT_INSTANCES=1 in ComponentRegistry.h
-this macro can be increased to allow more instances per codec
Signed-off-by: Daniel Charles <dcharles@ti.com>
Originally from: https://partner.source.android.com/g/#change,1132
Fixed many errors and cleaned up the code a bit.
Fixed potential free of unintialized memory
|
|
bug 2127768
OMX Component was not reseting first_buff flag when flushing output buffers.
Whenever the IL client does not flush input buffers the error reported is
present. Since the EOS has been resolved at the SN level, output buffer
can reset the firt_buff safely.
Originally from: https://partner.source.android.com/g/#change,1151
|
|
because the TI decoder has a requirement that the max dimensions
be multiples of 16
Originally from: https://partner.source.android.com/g/#change,1107
|
|
Originally from: https://partner.source.android.com/g/#change,987
|
|
Originally from: https://partner.source.android.com/g/#change,1109
|
|
It adds a mutex around the global allocation list used by the component,
and adds the required padding around the LCML config buffer.
Signed-off-by: Daniel Charles <dcharles@ti.com>
Originally from: https://partner.source.android.com/g/#change,1103
|
|
After this change, we only support the following hardware codecs:
- encoders: avc, m4v, h263, aac, amrnb, amrwb, jpeg
- decoders: avc, m4v
|
|
Originally from: https://partner.source.android.com/g/#change,1076
|
|
return all the buffers when the state transition from
executing to idle is completed.
bug 2055525
Originally from: https://partner.source.android.com/g/#change,1056
|
|
The hang happens when a reset command (state transiton from executing
-> idle) is issue while a port reconfiguration is going on.
Originally from: https://partner.source.android.com/g/#change,1073
Fixed a missing open brace and coding style.
|
|
Originally from: https://partner.source.android.com/g/#change,1057
Modified the hardcoded constant to use the MACRO instead.
|
|
|
|
from 4 to 6. 6 is the number of buffers actually being
supported already in the kernel.
Originally from: https://partner.source.android.com/g/#change,928
Removed all changes to the LOGE so that we keep the failure log
messages for now.
|
|
Fix playback hang issues by flushing frameparam memory when EOS flag is set
Originally from: https://partner.source.android.com/g/#change,933
|
|
bug 2055532
Originally from: https://partner.source.android.com/g/#change,932
Added change to use MP3D_OUTPUT_PORT and MP3D_INPUT_PORT, instead of OMX_DirOuptut and NULL,
based on review suggestions.
|
|
EOS to be propogated to PV.
Originally from https://partner.source.android.com/g/#change,860
Appears other OMX decoder components have similar issues
Thus, more fixes are expected to be coming soon.
|