Topology2 sched domain support #9959
Conversation
lgirdwood
left a comment
lgirdwood
left a comment
There was a problem hiding this comment.
LGTM, needs some comments. Btw, can you link kernel PR here too. Thanks !
| num_input_audio_formats 415 | ||
| num_output_audio_formats 416 | ||
| no_wname_in_kcontrol_name 417 | ||
| sched_domain 418 |
jsarha
force-pushed
the
topology2_sched_domain_support
branch
from
d3532ce to
906fff9
Compare
jsarha
force-pushed
the
topology2_sched_domain_support
branch
from
906fff9 to
ea09287
Compare
|
I finally got the DP thread to actually appear and work. First some topology language difficulties (little error somewhere but no error, and no attribute), and then finally hitting an error with gain module DP processing. The DP core needs a module that uses sink and source APIs (e.g. module process method is implemented and not only process_audio_stream, like in gain module). Thus this version now uses SRC module as it supports sink and source APIs. I've tested that it works and the DP processing thread is actually created on core 2 when audio is played to hw:0,2. |
kv2019i
left a comment
kv2019i
left a comment
There was a problem hiding this comment.
Looks ok. I was wondering about the duplicated tplg logic, please check inline.
| } | ||
| ] | ||
| } | ||
| } |
There was a problem hiding this comment.
I wonder if this could reuse nocodec more and just override the schedule. nocodec is one of the more complex so cloning it completely is not ideal for maintainance. For BT we have a scaled down nocodec that is easier to maintain.
DId @jsarha you consider this?
There was a problem hiding this comment.
I considered, but I could not figure out a way to do it without adding even more variables and even more IncludeByKey -sections to cavs-nocodec.conf. I feelt nocodec topology is already horrible enough to read. But yes. For this particular purpose it could actually make sense even if I do not really like the way we currently use the IncludeByKey -sections at the topology top level.
But eventually we need a major restructure to the whole topology hierarchy. More top-level files and less variation in each of them. The first in my kill list is PASSTHROUGH variable and the numerous IncludeByKey-sections sensitive to it.
But yes, I'll add one more variable and couple of more IncludeByKey-sections.
Add "scheduler_domain" attribute to all widgets. The allowed values are "LL", "DP" and "default". Signed-off-by: Jyri Sarha <jyri.sarha@linux.intel.com>
This commit creates three new topologies, sof-mtl-nocodec-dp-test.tplg, sof-lnl-nocodec-dp-test.tplg, and sof-ptl-nocodec-dp-test.tplg. They are otherwise the same as the corresponding standard nocodec topologies, but both the src.11.1 on SSP2_Playback and src.5.1 on SSP2 Capture have scheduler_domain attribute set to "DP". Signed-off-by: Jyri Sarha <jyri.sarha@linux.intel.com>
jsarha
force-pushed
the
topology2_sched_domain_support
branch
from
ea09287 to
f0195ef
Compare
|
@kv2019i in the end I survived without IncludeByKey-sections. |
3 participants
NOTE: thesofproject/linux#5382 is needed before this PR does anything.
The PR defines a new topology2 attribute for any widget. With the scheduler_domain attribute one can select if the component defined by the widget node is execute in Low Latency thread or in Data Processing mode. The PR also adds new test topologies for testing the DP mode.