scx_lavd detects wrong cpu configuration

[tr1xem]

May 16 11:13:15 am am cachyos scx_loader[342185]: 05:43:15 am am [INFO] CPU pref order in performance mode: [12, 13, 14, 15, 16, 17, 18, 19, 0, 2, 4, 6, 8, 10, 1, 3, 5, 7, 9, 11]
May 16 11:13:15 am am cachyos scx_loader[342185]: 05:43:15 am am [INFO] CPU pref order in powersave mode: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19]

Using scx_lavd I have i7-13650hx
According my cpu config 0-12 are performance core aka high power core and powersave use them is it not weird??
while performance mode use 12 - 19 which are low power threads aka efficient cores

I have i7-13650hx

CPU NODE SOCKET CORE L1d:L1i:L2:L3 ONLINE    MAXMHZ   MINMHZ       MHZ
  0    0      0    0 0:0:0:0          yes 2600.0000 800.0000 2600.0000
  1    0      0    0 0:0:0:0          yes 2600.0000 800.0000 2600.0259
  2    0      0    1 4:4:1:0          yes 2600.0000 800.0000 2600.0000
  3    0      0    1 4:4:1:0          yes 2600.0000 800.0000 2600.0000
  4    0      0    2 8:8:2:0          yes 2600.0000 800.0000 2136.8860
  5    0      0    2 8:8:2:0          yes 2600.0000 800.0000 1128.5850
  6    0      0    3 12:12:3:0        yes 2600.0000 800.0000 1599.2610
  7    0      0    3 12:12:3:0        yes 2600.0000 800.0000  800.0000
  8    0      0    4 16:16:4:0        yes 2600.0000 800.0000  948.2410
  9    0      0    4 16:16:4:0        yes 2600.0000 800.0000  800.0000
 10    0      0    5 20:20:5:0        yes 2600.0000 800.0000 2482.2520
 11    0      0    5 20:20:5:0        yes 2600.0000 800.0000  800.0000
 12    0      0    6 24:24:6:0        yes 1900.0000 800.0000 1900.6110
 13    0      0    7 25:25:6:0        yes 1900.0000 800.0000  800.0000
 14    0      0    8 26:26:6:0        yes 1900.0000 800.0000 1899.9611
 15    0      0    9 27:27:6:0        yes 1900.0000 800.0000 1899.7540
 16    0      0   10 28:28:7:0        yes 1900.0000 800.0000  800.0000
 17    0      0   11 29:29:7:0        yes 1900.0000 800.0000 1359.9220
 18    0      0   12 30:30:7:0        yes 1900.0000 800.0000  800.0000
 19    0      0   13 31:31:7:0        yes 1900.0000 800.0000 1555.4290

my lstopo

[multics69]

Thanks @tr1xem for reporting this and providing info (over Discord). Will take a look!

[tr1xem]

Looking forward to get it fixed Always ready for any kind of help. Thanks

[multics69]

@tr1xem -- PR #1903 should fix your problem. Please test it when you have some spare cycles.

[tr1xem]

@multics69
Dont really think it mitigated issue "completely".

05:08:28 [INFO] Autopilot mode is enabled.
05:08:28 [INFO] Opts {
    autopilot: true,
    autopower: false,
    performance: false,
    powersave: false,
    balanced: false,
    slice_max_us: 5000,
    slice_min_us: 300,
    cpu_pref_order: "",
    no_futex_boost: false,
    no_preemption: false,
    no_wake_sync: false,
    no_core_compaction: false,
    no_freq_scaling: false,
    stats: None,
    monitor: None,
    monitor_sched_samples: None,
    verbose: 0,
    version: false,
    help_stats: false,
}
05:08:28 [INFO] CPU pref order in performance mode: [0, 2, 4, 6, 8, 10, 12, 13, 14, 15, 16, 17, 18, 19, 1, 3, 5, 7, 9, 11]
05:08:28 [INFO] CPU pref order in powersave mode: [12, 13, 14, 15, 16, 17, 18, 19, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11]
05:08:28 [INFO] scx_lavd scheduler is initialized (build ID: 1.0.12-g2c3b186d x86_64-unknown-linux-gnu)
05:08:28 [INFO] scx_lavd scheduler starts running.

As you could see 1 3 5 7 8 11 are at last in performance mode they should be in front as they are part of pcore too. However Powersave mode's cpu pref are fine

Also this:
Not a error just a warning

warning: value assigned to `raw_capacity` is never read
   --> rust/scx_utils/src/topology.rs:606:13
    |
606 |     let mut raw_capacity = 0;
    |             ^^^^^^^^^^^^
    |
    = help: maybe it is overwritten before being read?
    = note: `#[warn(unused_assignments)]` on by default

warning: `scx_utils` (lib) generated 1 warning

[multics69]

@tr1xem -- Hmm, The only thing that I can imagine is that /sys/devices/system/cpu/cpu*/acpi_cppc/highest_perf changes dynamically depending on the load. That was not the case on my AMD laptop and Intel RaptorLake desktop. When you see the odd results, could you share the output of cat /sys/devices/system/cpu/cpu*/acpi_cppc/highest_perf and cat /sys/devices/system/cpu/cpu*/cpufreq/cpuinfo_max_freq?

[tr1xem]

@multics69
In idle
Highest Perf for all is : 225
Max Freq for P Cores : 2.6 ghz
Max Freq for E Cores : 1.9 ghz

Under Load Condtion (turbo boost on) some weird output
Cpu0 - Cpu3 4.7ghz
cpu4 - cpu7 4.9ghz
cpu8 - cpu 11 4.7ghz
cpu 13 - cpu19 3.6 ghz

highest pref is same for all still

[multics69]

05:08:28 [INFO] CPU pref order in performance mode: [0, 2, 4, 6, 8, 10, 12, 13, 14, 15, 16, 17, 18, 19, 1, 3, 5, 7, 9, 11]

@tr1xem - -I double-checked the output for performance mode. I found it works as designed: physical P-CPUs first 0, 2, 4, 6, 8, 10, then physical E-CPUs 12, 13, 14, 15, 16, 17, 18, 19, finally SMT-siblings of P-CPUs 1, 3, 5, 7, 9, 11. Cconsidering that an SMT sibling can deliver around 50-60% empirically, so this design was chosen.

Under Load Condtion (turbo boost on) some weird output
Cpu0 - Cpu3 4.7ghz
cpu4 - cpu7 4.9ghz
cpu8 - cpu 11 4.7ghz
cpu 13 - cpu19 3.6 GHz

It is weird in the sense that the cpuinfo_max_freq is updating dynamically. But the P-CPUs have higher frequencies than the E-CPUs. So it should be okay. I guess your cpu4-7 seems turbo-boostable.

So, it works as designed. I will close the issue for now.
Of course, there is room for improvement. If you have any suggestion, please let me know through github issue or discord.
Thanks!

[tr1xem]

The performance is better already pcores are getting turbo boost first and everything seems pretty stable now
The latency are lower and fps are higher in my testing.
Thankyou for Lavd :>