Tools: Topology2: Change in capture gain curve_duration to 50 ms

[singalsu]

This change increases the ramp duration from 20 ms to 50 ms. It lowers the peak load of peak volume component due to longer same gain value blocks. The internal update rate for gain becomes 250 us instead of 125 us. The longer fade-in ramp also conceals better possible analog capture start transients.

This changes for 4ch capture for gain.11.1 in sof-hda-generic-4ch.tplg from

CPU_PEAK(MAX) = 21.95
PEAK(MAX)/AVG(AVG) = 7.51

to

CPU_PEAK(MAX) = 9.07
PEAK(MAX)/AVG(AVG) = 3.12

[btian1]

change to 40/50/60ms have same result for peak ramp results, better to have a check for the result.

[lgirdwood]

@marcinszkudlinski @mwasko fyi - good improvement on MCPS

[lgirdwood]

@wszypelt @lrudyX fyi - not sure what volume gain curve duration you guys are using in CI today.

[btian1]

@singalsu @lgirdwood
https://sof-ci.ostc.intel.com/#/result/planresultdetail/32024

seems CI still failed, either change CI pass reference or revert this patch.

[mszleszy]

@lgirdwood tests uses different curve duration, and according to my knowledge it should not be static as curve length is being defined in InitInstance ipc or LargeConfigSet message of Gain module
example for 0.1s of fade in:

[singalsu]

@singalsu @lgirdwood https://sof-ci.ostc.intel.com/#/result/planresultdetail/32024

seems CI still failed, either change CI pass reference or revert this patch.

@btian1 If you mean issue #8238 the bug is in CI, not in topology or gain component. Alsabat or using alsabat from sof-test should have a way to ignore from start the ramp time.

[singalsu]

@lgirdwood tests uses different curve duration, and according to my knowledge it should not be static as curve length is being defined in InitInstance ipc or LargeConfigSet message of Gain module example for 0.1s of fade in:

@mszleszy Yes, the gain component follows the curve_duration and curve_type request from init() IPC. But currently the curve duration and curve type from successive volume commands is ignored. Also we can't have different curve per channel.

[mszleszy]

@singalsu as far as i know setting up different configs per channel was supported up to cavs2.5 It's not required any longer but the structure stayed the same for back compatibility.
I can speak only form fw QA perspective that PeakVol and Gain configuration worked pretty well so far, so any new fail on those modules looks like a regression. But if necessary we can do a deep dive for this case

[singalsu]

@singalsu as far as i know setting up different configs per channel was supported up to cavs2.5 It's not required any longer but the structure stayed the same for back compatibility. I can speak only form fw QA perspective that PeakVol and Gain configuration worked pretty well so far, so any new fail on those modules looks like a regression. But if necessary we can do a deep dive for this case

Yep, we should not attempt to tame the peak MCPS with ways those cause regressions.