Bug: Wrong power measurement on iPad 2020

[hamjin]

Version: 2.3.0
Device: iPad 8
iPad OS Version: 16.5
SoC: Apple A12

Steps to reproduce

Open the app first. Then select one of the project and run.
I'm sure that the device is not charging or doing anything else.

Expected behaviour

Show correct power.

Actual behaviour

Always showing wrong power like below.

[junjie1475]

First, thanks for opening up an issue. The power measurement is from apple's CLPC(Close loop performance controller)which I believe only appear on SoC >A13. I used this to get high resolution per process energy consumption; so under current circumstances, there is no way to workaround with this. Maybe there are other ways to get this but I am sure it wouldn't be as accurate as the current approach.

[hamjin]

IOService.txt


It's clear that Apple A12 has CLPC

[junjie1475]

From XNU source code(Version 8792.81.2), the power energy counter code is gated behind macro HAS_CPU_DPE_COUNTER, and apple had only published the header(see https://github.com/apple-oss-distributions/xnu/blob/5c2921b07a2480ab43ec66f5b9e41cb872bc554f/pexpert/pexpert/arm64/H13.h#L50) for H13(A14) which defineds that macro; I don't really know if there is another way to get those header file(after Version 8792.81.2).

[hamjin]

Can you can provide a power calibration interface like GPU GFLOPS? I think that helps with debugging.

[junjie1475]

Can you can provide a power calibration interface like GPU GFLOPS? I think that helps with debugging.

If you mean high resolution sampling, see my another repo https://github.com/junjie1475/ios_power_log

[TomyLemon]

Can you can provide a power calibration interface like GPU GFLOPS? I think that helps with debugging.

Do you mean a power measure app with GUI and run in the background ？

[hamjin]

Can you can provide a power calibration interface like GPU GFLOPS? I think that helps with debugging.

If you mean high resolution sampling, see my another repo https://github.com/junjie1475/ios_power_log

I think it may be a wrong unit of measurement causing the power to show extremely low.

[hamjin]

Can you can provide a power calibration interface like GPU GFLOPS? I think that helps with debugging.

Do you mean a power measure app with GUI and run in the background ？

If you can, that's great.

[junjie1475]

Can you can provide a power calibration interface like GPU GFLOPS? I think that helps with debugging.

If you mean high resolution sampling, see my another repo https://github.com/junjie1475/ios_power_log

I think it may be a wrong unit of measurement causing the power to show extremely low.

From XNU documentation(see recount.md) the unit is nano-joule, but you can try to use ios_power_log to track the energy consumption/sec and see what cause the bug.

[junjie1475]

Can you can provide a power calibration interface like GPU GFLOPS? I think that helps with debugging.

Do you mean a power measure app with GUI and run in the background ？

If you can, that's great.

I don't think it's possible to track other process's info on iOS but I can integrate the graph into the App itself perhaps(Maybe through swift chart).

[hamjin]

Can you can provide a power calibration interface like GPU GFLOPS? I think that helps with debugging.

If you mean high resolution sampling, see my another repo https://github.com/junjie1475/ios_power_log

I think it may be a wrong unit of measurement causing the power to show extremely low.

From XNU documentation(see recount.md) the unit is nano-joule, but you can try to use ios_power_log to track the energy consumption/sec and see what cause the bug.

Sorry but I don't have a computer running macOS.

[junjie1475]

Can you can provide a power calibration interface like GPU GFLOPS? I think that helps with debugging.

If you mean high resolution sampling, see my another repo https://github.com/junjie1475/ios_power_log

I think it may be a wrong unit of measurement causing the power to show extremely low.

From XNU documentation(see recount.md) the unit is nano-joule, but you can try to use ios_power_log to track the energy consumption/sec and see what cause the bug.

Sorry but I don't have a computer running macOS.

You don't need to have one, as far as I know you can use sideloadly to inject the dylib and it is also possible to read console log through windows.

[TomyLemon]

Can you can provide a power calibration interface like GPU GFLOPS? I think that helps with debugging.

If you mean high resolution sampling, see my another repo https://github.com/junjie1475/ios_power_log

I think it may be a wrong unit of measurement causing the power to show extremely low.

In my opinion, it seems right. As you can see, the power measured by the app is slightly lower than the power measured by macOS powermetrics command.

This command shows the cpu power, while our app measures the spec2017 task power, so it will be a bit lower.

[asto18089]

Can you can provide a power calibration interface like GPU GFLOPS? I think that helps with debugging.

If you mean high resolution sampling, see my another repo https://github.com/junjie1475/ios_power_log

I think it may be a wrong unit of measurement causing the power to show extremely low.

In my opinion, it seems right. As you can see, the power measured by the app is slightly lower than the power measured by macOS powermetrics command. This command shows the cpu power, while our app measures the spec2017 task power, so it will be a bit lower.

Yes, but I believe he uses Apple A12 but not Apple M1, something could be different for that chip.

I agree with junjie that the difference could reside in differences between Apple A12's CLPC and Apple M1's CLPC, and in XNU opensource only M1's CLPC definition could be found, so a headache here......

[hamjin]

This is a section of my running 500.perlbench results (from rest to running), without power saving mode enabled

[junjie1475]

This is a section of my running 500.perlbench results (from rest to running), without power saving mode enabled

I am not very sure what is the power usage of the A12 device but it seems like power x100 would lead to the correct result.

[hamjin]

This is a section of my running 500.perlbench results (from rest to running), without power saving mode enabled

I am not very sure what is the power usage of the A12 device but it seems like power x100 would lead to the correct result.

Only one result from Anandtech's Spec 2006 has subitem power consumption and I don't know the gap.

[junjie1475]

Ok then, I will patch this fix when I have time.

…

On Sun, 4 Jun 2023 at 5:59 PM, hamjin ***@***.***> wrote: [image: 697799afd74fbe72f93ddd01caf06ed0] <https://user-images.githubusercontent.com/30927889/243166013-5a79d08a-c4bd-4655-acfe-5997a60dfb77.png> [image: image] <https://user-images.githubusercontent.com/30927889/243166045-6bcfe3f1-6840-4f8a-a2dd-a78a9abe4491.png> This is a section of my running 500.perlbench results (from rest to running), without power saving mode enabled I am not very sure what is the power usage of the A12 device but it seems like power x100 would lead to the correct result. [image: image] <https://user-images.githubusercontent.com/30927889/243168910-07037421-8ec5-4918-a012-8e933be76dec.png> Only one result from Anandtech's Spec 2006 has subitem power consumption and I don't know the gap. — Reply to this email directly, view it on GitHub <#3 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AUSI37B22TRVINESQ5HO7E3XJRL6PANCNFSM6AAAAAAYTNW6QA> . You are receiving this because you were assigned.Message ID: ***@***.***>

[hamjin]

Ok then, I will patch this fix when I have time.
…
On Sun, 4 Jun 2023 at 5:59 PM, hamjin @.> wrote: [image: 697799afd74fbe72f93ddd01caf06ed0] https://user-images.githubusercontent.com/30927889/243166013-5a79d08a-c4bd-4655-acfe-5997a60dfb77.png [image: image] https://user-images.githubusercontent.com/30927889/243166045-6bcfe3f1-6840-4f8a-a2dd-a78a9abe4491.png This is a section of my running 500.perlbench results (from rest to running), without power saving mode enabled I am not very sure what is the power usage of the A12 device but it seems like power x100 would lead to the correct result. [image: image] https://user-images.githubusercontent.com/30927889/243168910-07037421-8ec5-4918-a012-8e933be76dec.png Only one result from Anandtech's Spec 2006 has subitem power consumption and I don't know the gap. — Reply to this email directly, view it on GitHub <#3 (comment)>, or unsubscribe https://github.com/notifications/unsubscribe-auth/AUSI37B22TRVINESQ5HO7E3XJRL6PANCNFSM6AAAAAAYTNW6QA . You are receiving this because you were assigned.Message ID: @.>

Sorry, the power measured in the Geekbench 6 multi-core test does not support previous guesses. Either ri_penergy nj ("power" in the picture) or ri energy nj` ("power" in the picture) may be incorrect on the Apple A12 or have other reasons. Maybe we have to collect more information about this SoC.

[junjie1475]

Ok then, I will patch this fix when I have time.

…

On Sun, 4 Jun 2023 at 5:59 PM, hamjin @.> wrote: [image: 697799afd74fbe72f93ddd01caf06ed0] https://user-images.githubusercontent.com/30927889/243166013-5a79d08a-c4bd-4655-acfe-5997a60dfb77.png [image: image] https://user-images.githubusercontent.com/30927889/243166045-6bcfe3f1-6840-4f8a-a2dd-a78a9abe4491.png This is a section of my running 500.perlbench results (from rest to running), without power saving mode enabled I am not very sure what is the power usage of the A12 device but it seems like power x100 would lead to the correct result. [image: image] https://user-images.githubusercontent.com/30927889/243168910-07037421-8ec5-4918-a012-8e933be76dec.png Only one result from Anandtech's Spec 2006 has subitem power consumption and I don't know the gap. — Reply to this email directly, view it on GitHub <#3 (comment)>, or unsubscribe https://github.com/notifications/unsubscribe-auth/AUSI37B22TRVINESQ5HO7E3XJRL6PANCNFSM6AAAAAAYTNW6QA . You are receiving this because you were assigned.Message ID: @.>

Sorry, but power measurements on Geekbench 6 don't support previous guesses. Either ri_penergy nj ("power" in the picture) or ri energy nj` ("power" in the picture) may be incorrect on the Apple A12 or have other reasons. Maybe we have to collect more information about this SoC.

I think it is possible that those values don't have any meaning. I should look more into the XNU code to see if I can find more clues.

[hamjin]

Thanks again for your efforts, if help is needed I will do my best.