[Merged] Power Saver feature

Hex

@cyperghost in short SDL2 event handling is shit.

How a normal framework should handle key presses

The UI is rendered on the main thread. When a key is pressed, the OS should send an interrupt/signal to the framework. Framework checks which method handles key press and calls that method to handle key press. The method should return if it consumes the key press or not. If key press was not consumed then the framework passes it to its internal key press handler.

How SDL2 handles key press
SDL2 main thread check if user has pressed a key by SDL_pollevent or SDL_waitevent. When a user presses a key it is added to event queue. SDL_poll call checks if there is an event in event queue and returns with the first incoming event (null if no event). the main thread then handles the incoming event and renders the frame accordingly. Since we need lots of frame for smooth animations this has to be put in a timed loop. The timer decides which frame is being rendered.

The problem starts here. Now that you are in a while loop for rendering, what happens if no events are received? You just keep on looping doing nothing but rendering the screen which is not changing, thus consuming CPU.

SDL_Wait come to the rescue. SDL_wait waits for an event to occur and only then returns. so it is a blocking call. If an event doesn't occur, your main thread is not rendering. Hence the CPU savings. But what does SDL_wait do in the background? It just polls and sleeps in a loop till an event occurs, which is stupid.

cyperghost

@Hex Okay I understand. So it's a question of the framwork and the programmer (aloshi?) was forced to use this stupid polling events.

It's really a nice that you made this possible! And I hope that this fix will find it's place in ES development!

Nearby for info @interested:
Polling is a timed loop waiting for events - just stupid. You are doing this if you have really no knowledge how a process is working or the surrounded framework lacks of better techniques. Then you are using polling events! So you let the CPU do something (or let the CPU just waiting in background) and after 100µS you check an event (button press...) and it says "Something happend" or "Nothing happend". Then you can produce errors like bouncing ... make 2 steps instead of one.

The better way is to use interrupts. You use a trigger that stays permanent in memory and this trigger is waiting for an event. If the event has started than a vector jumps back to the programm and says "Something happend!" and that's always better for CPU usage but needs a bit more programming skills!

Think about 2 humans talking to each other.

The Poller:

Person 1: What's up?
Person 2: Nothing!
After 5 minutes
Person 1: What's up?
Person 2: Nothing!
After 5 minutes
Person 1: What's up?
Person 2: Nothing!
After 5 minutes
Person 1: What's up?
Person 2: You get me on the nerves?
Person 1: Really - what you want me to do?
Person 2: Get me out of this

The Interrupter

Person 1: If there is a need for smalltalk... then please tell me! (=set pointer to person 2)
5 min
5 min
5 min
Person 2: It's so boring (=vector back to Person 1 = Interrupt request or IRQ)
Person 1: What you want me to do?
Person 2: Get me out of this

;)

Hex

@cyperghost The programmer Alec "Aloshi" Lofquist, was probably coding on a PC and did not make things for small devices. Stock ES uses 1% at idle on my Linux laptop which equates to 60% on Pi 0.

~~Lets say ES is a worker and OS is his manager. ES is at his desk.~~ Never mind you just edited to add something like this :| And the 5mins is more like 1ms

cyperghost

@Hex
I just wanted to write something everbody can imagine! It's better to talk about a pointer that is setted to a fixed RAM cell and sends a vector back to the pointer with annother value? Of course we are talking about GHz so 1ms is really a long long time :)
Thank you for the technical background.

Sano

Optimizing is ALWAYS a good thing, even on the most powerful hardware.

I know most devs (except maybe linux/free software devs) nowadays think it will be cheaper and faster to just add more RAM or whatever, but this is wrong on so many levels...

As a sysadmin, I can swear I've seen so many poorly coded applications that I just want to flip a table over some devs head.
I have tons of examples if you want to laugh :)
Including some SQL request going down from 20min to 4s after simple optimization, or a soft doing 70000 iops (yup !) for a 3min long on-user-demand treatment that was optimized to a mere 15s (and 150iops usage).
God sometimes I hate my life...

Hex

Are your systems running on batteries or connected to a power supply? My battery life testing is quite simple. if i have to test different batteries I run Contra on NES emu. It has something like attract mode where the Cpu plays so i get comparison between different batteries or boost circuits.

With ES I switch screensaver off and just leave it on Systems view. This allows me to compare power savings over multiple ES builds.

Hex

I am switching to PS enabled by default and all changes are being finalized. Would the testers like to report any bugs or problems with the branch. One major problem is auto launch of screensaver doesnt work. Manual trigger works. I am looking into that and once that is stable the PR would be ready for merge

meleu

@Hex said in Testers needed :: Power Saver features :: PR #172:

One major problem is auto launch of screensaver doesnt work. Manual trigger works.

I like this behavior. Can we label it as a feature instead of a problem. :-)

Sano

Ok, sorry guys, was short on free time the few past days.

I installed the powersaver version of ES on a Pi3, and just... Wow !
We're talking about a night and day difference here.

As I previously stated, ES CPU usage when active was around 30 to 40%, and dropped to 3-6% only after the screensaver timeout (5min).

Now we have :
25% when active
after a few seconds only, it drastically drops to 0.3 %
Very impressive !

I'll let the option activated in order to test more in real conditions, but I can't wait the merge with the main branch.
Congratulations @Hex

Edit :
@Hex Is the game number on carousel tweak an addition of yours too ? It may be me, but I didn't see it before using your version.
Anyway, this is brilliant.

felleg

@meleu said in Testers needed :: Power Saver features :: PR #172:

@Hex said in Testers needed :: Power Saver features :: PR #172:

One major problem is auto launch of screensaver doesnt work. Manual trigger works.

I like this behavior. Can we label it as a feature instead of a problem. :-)

This is a problem if you want to avoid screen burn-in. I'd much rather have the screensaver be able to start automatically, although, don't get me wrong, being able to start it manually is great in its own right.

cyperghost

@felleg Please explain "Screen Burn In" on LCD/TFT

Hex

@meleu Hehe, i dont take any "prisoners of war". Its bad when bugs start being labelled as features.

@Sano Thank you. I dont understand what you mean by game number on carousel? Are you taking about the number of games in a system shown. That is available in the main branch since a long time.

@felleg Worry not, the screensavers video/dim/blank are working as expected with the PS mod. The logic is becoming very complex on the other hand.

@cyperghost there are some who use it on plasma or retro TV sets, might be for them?

Sano

@Sano Thank you. I dont understand what you mean by game number on carousel? Are you taking about the number of games in a system shown. That is available in the main branch since a long time.

Yes, but I realized on your version that the number doesn't appear automatically after a delay. I have to press down direction for it to appear.
I like it, it's more convenient for me to not have this information unless I want it :)

Hex

@Sano Again, bugs are not features. It might be okay for you but for someone who doesnt know about this behaviour will be perplexed and complain on forums.

Also what system are you getting this behaviour in?

Sano

All of them :) (Mame/FBA/NES/SNES/SMS/Megadrive/Ports/Retropie)
I have to press a button (not only the down direction after some tests) for the number of games to appear.

Hex

System as in Pi1/2/3 Linux PC??

Sano

Sorry.
RPi3, retropie 4.2.8, 3A Aukru power supply.

cyperghost

@Hex Okay... That's a point :) So ScreenBurn might be an issue.

Other question. Can you help me out please? I wanted to melt your branch together with @pjft but I wasn't able to do this ... The issue is posted here

I've also uploaded the binary you've provided for RPI1/0 build to my github

https://github.com/crcerror/emulationstation-binary-archive

Hex

@Sano are you compiling the binary?

Sano

@Hex compiled via @meleu script, yes !