Same FPS, slower games?

[Iulius]

i allready tried to find it, but dont had success.

its not anywhere near the changes of 3545, must be somewhere deeper inside.

that also makes sense, as not even nolan(the guy who made 3545) really knows about these effects. ( compare with his statement at issue 1050)


to answer your question : no, sorry. im not able to fix it.

you will have to wait for somebody who knows whats going on deep in main core AND is willing to fix the framelimiter


hopefully the whole timing will be fixed instead, of course, but i think its unlikely to happen.

seems like the devs want the framecounter to show 2-3 times more fps even if the game is not really running that fast.

[CacoFFF]

Sounds interesting, on the other side, i tried changing the watchdog back to 15 in r3548.
Brawl got fast again but i noticed something interesting: The counter in the right of xx/xx (which usually is ~60) adjusts itself to 30 when fps goes between 20 and 40 and then goes back to 60 when fps gets better. These changes happen even in the middle of a fight.

It's like the game has two modes, when the framerate drops to less that 35, the game starts updating every 1/25 or 1/30 (depending of region) seconds instead of 1/50 - 1/60. Game will start slowing down again once fps drops from 20 and i doubt it has a 1/15 mode, my cpu isn't that weak to be able to test that.

Want a patch to see for yourself?

[Iulius]

Quote:It's like the game has two modes, when the framerate drops to less that 35, the game starts updating every 1/25 or 1/30 (depending of region) seconds instead of 1/50 - 1/60

would be nice, but is wrong.

look in core/core.cpp function Callback_VideoCopiedToXFB(), where framelimiter is.

there you can see Jpetersons function and his comment :

Quote:Correct the FPS/VPS rate for temporary CPU-GPU timing variations....depending on which it's closest to.

in other words : it just looks how many fps you got and calculates what realspeed that MIGHT be related to.

but that has NOTHING to do with the real speed nor does it change any speed.

its just a try to display the right realspeed depending on the speed you have at the moment.

for example : if a game runs with 20fps on pal, it will most likely show 20/25fps, no matter if real gamespeed is 50.

Quote:Brawl got fast again

i really wonder why... melee does not act like that.

i dont got brawl, so i cant test that.

[CacoFFF]

I see, so the fps stats jpeterson added is plain arbitrary.

Then Brawl indeed has it's own timing stuff, as far as i've heard, Tales of Symphonia 2 and Metroid Prime 2 behave this way as well. I you have any of those games (i don't) you can try it out.

[Iulius]

i only got melee, which reacts like that with pre 3545 and low watchdog(off works best), so i thought it would be same for brawl and co, but it seems its not.

[gcp111]

Hmm. I guess we'll have to wait then.

Are there really no frame start/end functions? o_O

[Iulius]

stupid me...

i tried the whole time to find a way in the core with 3545, but just realized that this "way" was removed with 3545 !

(obviously it was, cause update now comes from video xfb update)

so i need to find a place for the framelimiter in PRE 3545 that still works after 3545.

or somebody who knows about it just tells me, but i think either the devs dont read here or they like to see me struggleling to find that stupid point in the code that says :
"hi, im the cpu and i want the gpu to render a frame"

a pity i cant just search for that

[CacoFFF]

I think i'll wait for the end of the day to release x86-x64 builds with the watchdog patch applied so ppl can test it, devs have been making a lot of commits lately.
This way we'll be able to see if framelimiter works properly and if there are other games behaving like Brawl.

[Willysonicteam]

The mario galaxy in the new versions (example 3590)run to 30fps,and the game is also slow.In the older versions run in more fps and have the same speed of the new verions.
is dificult for undestand,tested the version 3542 and compare.

[CacoFFF]

Test anything prior to 3397, that commit was the one that changed the watchdog value.