Telemetry: change performance timing method to ccount #9361
Conversation
tobonex
changed the title
tobonex
force-pushed
the
telemetry_perf_count_change
branch
from
9f88050
to
d0964dc
Compare
tobonex
requested review from
pblaszko,
marcinszkudlinski,
dbaluta,
LaurentiuM1234,
plbossart,
mmaka1,
lbetlej and
kv2019i
as code owners
src/schedule/zephyr_ll.c
Outdated
|
|
||
| telemetry_update(begin_stamp, current_stamp); | ||
| #endif | ||
|
|
There was a problem hiding this comment.
So this is called more often after the patch, I guess this is intended in the patch.
There was a problem hiding this comment.
Hm, I intended this to be more aligned with performance overlay... but this does seem to update the values more often which seems unnecessary. Not sure how it should work. I could delete this one commit for now, as this not a main focus of this PR.
| @@ -110,3 +110,4 @@ CONFIG_COMP_GOOGLE_RTC_AUDIO_PROCESSING=y | |||
| CONFIG_COMP_STUBS=y | |||
There was a problem hiding this comment.
The git commit prefix is wrong, this patch applies to one Intel platform only. And what about other Intel SOF targets? Shouldn't we have same timer source for all Intel platforms?
tobonex
force-pushed
the
telemetry_perf_count_change
branch
from
d0964dc
to
c3f6f5b
Compare
Adds abstract function to change time source for performance measurement via config. Signed-off-by: Tobiasz Dryjanski <tobiaszx.dryjanski@intel.com>
Change performance counter for telemetry to abstract one. This enables choice of time source. Signed-off-by: Tobiasz Dryjanski <tobiaszx.dryjanski@intel.com>
Change performance counters to use abstract function to use multiple time sources via config. Signed-off-by: Tobiasz Dryjanski <tobiaszx.dryjanski@intel.com>
Use ccount for performance measurement for mtl to have a better comparison with reference firmware. Signed-off-by: Tobiasz Dryjanski <tobiaszx.dryjanski@intel.com>
Use ccount for performance measurement for lnl to have a better comparison with reference firmware. Signed-off-by: Tobiasz Dryjanski <tobiaszx.dryjanski@intel.com>
tobonex
force-pushed
the
telemetry_perf_count_change
branch
from
c3f6f5b
to
be2738d
Compare
|
@kv2019i good for you now ? |
There was a problem hiding this comment.
Ok, let's proceed. The git commit prefixes are still wrong, but this is a problem with other PRs as well, so I won't block because of this. E.g. for the ace15 config file, the history looks like this:
55ecc19ab6e5 app: intel_adsp_ace15_mtpm.conf: set DRC as built-in
cfd27fbde0e1 src: convert to a loadable module
9f480375057e llext: automatically calculate module addresses
Only first one is correct, so it's not just this PR.
But oh well, you did add the board mentions on the summary lines, so this is already better.
For future, git commits should have a module prefix and this should make clear what is changed in a commit:
[area]: [summary of change]
[Commit message body (must be non-empty)]
Signed-off-by: [Your Full Name] <[your.email@address]>
When it comes to performance, to compare Sof with its reference we should use the same method of timing. Changing timing source to ccount for telemetry and performance monitor.