dp: Eearliest deadline first scheduling policy #8262
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
marcinszkudlinski
requested review from
pblaszko,
dbaluta,
LaurentiuM1234,
ranj063,
jxstelter,
singalsu,
abonislawski,
lgirdwood,
plbossart,
mmaka1,
lbetlej and
kv2019i
as code owners
marcinszkudlinski
added
MTL
lgirdwood
reviewed
Sep 26, 2023
lyakh
approved these changes
Sep 27, 2023
For LL modules it is very straightforward - all LL modules are called one by one in every LL pipeline tick, so module's period is always equal to LL pipeline period For DP modules a period has different meaning. As DP is called "on demand" - whenever it is ready, it may not be scheduled at every LL tick (and usually is not) A period meaning for DP is "a time when the module must finish its job and provide data to next module in pipeline or the data flow will be corrupted (a glitch)" In common case the period is calculation is based on module's OBS - if module provides OBS bytes of data in each cycle, it MUST finish at least before the next module drains the buffer In some special cases - modules producing a variable bitrate (like MPEG) or not producing audio data at all (phrase detector, speech recognition, etc.) - the module may set the period itself during prepare operation This commit adds calculation of DP period. This value will be used in following commit as a base for deadlines calculation in EDF Signed-off-by: Marcin Szkudlinski <marcin.szkudlinski@intel.com>
Earliest deadline first scheduling added, based on module period Detailed description in zephyr_dp_schedule.c Signed-off-by: Marcin Szkudlinski <marcin.szkudlinski@intel.com>
marcinszkudlinski
force-pushed
the
dp-edf-upstream
branch
from
a3bcdc8
to
b09ca6d
Compare
lgirdwood
approved these changes
Sep 27, 2023
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR introduces a Earlier Deadline First scheduling policy for DP processing
currently there's a limitation - DP module must be surrounded by LL modules.
it simplifies algorithm - there's no need to browse through DP chains calculating
deadlines for each module in function of all modules execution status.
Now is simple - modules deadline is its start + tick time.
example:
Lets assume we do have a pipeline:
LL1 -> DP1 -> LL2 -> DP2 -> LL3 -> DP3 -> LL4
all LLs starts in 1ms tick
for simplification lets assume
DP1 - starts every 1ms, needs 0.5ms to finish processing
DP2 - starts every 2ms, needs 0.6ms to finish processing
DP3 - starts every 10ms, needs 0.3ms to finish processing
TICK0
only LL1 is ready to run
LL1 processing (producing data chunk for DP1)
TICK1
LL1 is ready to run
DP1 is ready tu run (has data from LL1) set deadline to TICK2
LL1 processing (producing second data chunk for DP1)
DP1 processing for 0.5ms (consuming first data chunk, producing data chunk for LL2)
CPU is idle for 0.5ms
TICK2
LL1 is ready to run
DP1 is ready tu run set deadline to TICK3
LL2 is ready to run
LL1 processing (producing data chunk for DP1)
LL2 processing (producing 50% data chunk for DP2)
DP1 processing for 0.5ms (producing data chunk for LL2)
CPU is idle for 0.5ms
TICK3
LL1 is ready to run
DP1 is ready tu run set deadline to TICK4
LL2 is ready to run
LL1 processing (producing data chunk for DP1)
LL2 processing (producing rest of data chunk for DP2)
DP1 processing for 0.5ms (producing data chunk for LL2)
CPU is idle for 0.5ms
TICK4
LL1 is ready to run
DP1 is ready tu run set deadline to TICK5
LL2 is ready to run
DP2 is ready to run set deadline to TICK6
LL1 processing (producing data chunk for DP1)
LL2 processing (producing 50% of second data chunk for DP2)
DP1 processing for 0.5ms (producing data chunk for LL2)
DP2 processing for 0.5ms (no data produced as DP2 has 0.1ms to go)
100% CPU used
TICK5
LL1 is ready to run
DP1 is ready tu run set deadline to TICK6
LL2 is ready to run
DP2 is in progress, deadline is set to TICK6
LL1 processing (producing data chunk for DP1)
LL2 processing (producing rest of second data chunk for DP2)
DP1 processing for 0.5ms (producing data chunk for LL2)
DP2 processing for 0.1ms (producing TWO data chunks for LL3)
CPU is idle for 0.4ms (60% used)
TICK6
LL1 is ready to run
DP1 is ready tu run set deadline to TICK7
LL2 is ready to run
DP2 is ready to run set deadline to TICK8
LL3 is ready to run
LL1 processing (producing data chunk for DP1)
LL2 processing (producing 50% of second data chunk for DP2)
LL3 processing (producing 10% of first data chunk for DP3)
DP1 processing for 0.5ms (producing data chunk for LL2)
DP2 processing for 0.5ms (no data produced as DP2 has 0.1ms to go)
100% CPU used
(........ 9 more cycles - LL3 procuces 100% of data for DP3......)
TICK15
LL1 is ready to run
DP1 is ready tu run set deadline to TICK16
LL2 is ready to run
DP2 is ready to run set deadline to TICK17
LL3 is ready to run
DP3 is ready to run set deadline to TICK25
LL1 processing (producing data chunk for DP1)
LL2 processing (producing 50% of data chunk for DP2)
LL3 processing (producing 10% of second data chunk for DP3)
DP1 processing for 0.5ms (producing data chunk for LL2)
DP2 processing for 0.5ms (no data produced as DP2 has 0.1ms to go)
100% CPU used -
!!! note that DP3 is ready but has no chance to get CPU in this cycle
TICK16
LL1 is ready to run set deadline to TICK17
DP1 is ready tu run
LL2 is ready to run
DP2 is in progress, deadline is set to TICK17
LL3 is ready to run
DP3 is in progress, deadline is set to TICK25
LL1 processing (producing data chunk for DP1)
LL2 processing (producing rest of data chunk for DP2)
LL3 processing (producing 10% of second data chunk for DP3)
DP1 processing for 0.5ms (producing data chunk for LL2)
DP2 processing for 0.1ms (producing data)
DP3 processing for 0.2ms (producing 10 data chunks for LL4)
90% CPU used
TICK17
LL1 is ready to run
DP1 is ready tu run
LL2 is ready to run
DP2 is ready to run
LL3 is ready to run
LL4 is ready to run
!! NOTE that DP3 is not ready - it will be ready again in TICK25
LL1 processing (producing data chunk for DP1)
LL2 processing (producing rest of data chunk for DP2)
LL3 processing (producing next 10% of second data chunk for DP3)
LL4 processing (consuming 10% of data prepared by DP3)
DP1 processing for 0.5ms (producing data chunk for LL2)
DP2 processing for 0.5ms (no data produced as DP2 has 0.1ms to go)
100% CPU used
Now - pipeline is in stable state, CPU used almost in 100% (it would be 100% if DP3
needed 1.2ms for processing - but the example would be too complicated)