[ADL] ITBM support for Hybrid platforms (out-of-tree)

Please find the following patches in the above external link

0001-x86-sched-Decrease-further-the-priorities-of-SMT-sibl.itmt
0002-sched-topology-Introduce-sched_group-flags.itmt
0003-sched-fair-Optimize-checking-for-group_asym_packing.itmt
0004-sched-fair-Provide-update_sg_lb_stats-with-sched-doma.itmt
0005-sched-fair-Carve-out-logic-to-mark-a-group-for-asymme.itmt
0006-sched-fair-Consider-SMT-in-ASYM_PACKING-load-balance.itmt
0007-sched-fair-Check-asym_packing-priority-on-the-wakeup-.itmt

0001-x86-sched-Decrease-further-the-priorities-of-SMT-sibl.itmt --> 183b8ec38f1ec6c1f8419375303bf1d09a2b8369
0002-sched-topology-Introduce-sched_group-flags.itmt --> 16d364ba6ef2aa59b409df70682770f3ed23f7c0
0003-sched-fair-Optimize-checking-for-group_asym_packing.itmt --> 6025643596895695956c27119c6b0bfa40d8035b
0004-sched-fair-Provide-update_sg_lb_stats-with-sched-doma.itmt --> c0d14b57fe0c11b65ce8a1a4a58a48f3f324ca0f
0005-sched-fair-Carve-out-logic-to-mark-a-group-for-asymme.itmt --> aafc917a3c31dcc76cb0279cd7617dda11b15f2a
0006-sched-fair-Consider-SMT-in-ASYM_PACKING-load-balance.itmt --> 4006a72bdd93b1ffedc2bd8646dee18c822a2c26
0007-sched-fair-Check-asym_packing-priority-on-the-wakeup-.itmt --> drop it

A test kernel based on Ubuntu-intel-5.13-5.13.0-1008.8 with the following four patches is available @ https://people.canonical.com/~alexhung/LP1943549/

@chaoqin,

I pushed the patches to https://code.launchpad.net/~alexhung/+git/focal/+ref/iotg-adl/1943549 with comments above.

Please have a look and keep me posted if there are improvements or fixes to be made

Regarding the testcases for Hybrid scheduling, we run the following workloads for ADL-S/P:
1. Geekbench
2. Speedometer
3. Netperf
4. SpecInt
5. OpenVino (benchmark app) with Resnet50

All of these are standard Linux applications, hence you can use them for your testing.

We have used Geekbench extensively for this purpose. The sample command which we have used is:
./geekbench_x86_64 --no-upload --multi-core --cpu-workers <no of cores to be used> --iterations <no of iteration>

Strategy:
For ADL-S 881 SKU, we have 8 big cores and 8 atom config, where we have started geekbench for first 8 cores and observed the work distribution among multiple threads, we can use top or turbostat for this analysis. Second experiment is to execute geekbench with 16 cores(cpu workers) and then third experiment with 24 cores.

We can also manually disable and enable cores just to make sure if the results are same as the geekbench’s internal core pinning.

WARNING: When running on sensitive (e.g., pre-release or otherwise confidential)
platforms, always keep the system unplugged from the internet. Do not trust a
command line feature to protect sensitive information, as it introduces
more opportunity for human error or a benchmark bug to cause a leak.