[BUG] Increased response delay between 0.9.6 and 0.9.7

[tsukiyomaru0]

Welp, it has happened. For note, here are some info.
Intel Dual Core 3+3 ghz, 3 gb ram, GeForce 9, Windows XP, wired PS2 USB Controller.

In 0.9.6 the respone delay between a press and the action reproduced in the emulator was 0.5 seconds or less (considerably less in other emulators such as VBA, or in actual games like Blue Wish Resurrection Plus (free SHMUP).)

However, as of 0.9.7, the response delay has increased. It lies now between 1 and 2 seconds. This is a terrible desync, specially considering the driver I use for this controller is a very light one called USB Network Gamepad (Why this one? Because Rumble and accurate maping of Right and Left Analogs, where the generic USB Gamepad swaps D-Pad and Left Analog)

Please, look into it and fix this delay as soon as possible because, so far, games are unplayable with Gamepad on my end.

[[Unknown]]

You should try downloading some git builds in between to see where the problem for you started.

For example, I use a USB adapter for a PS1 controller, and last time I checked I was not seeing any crazy latency like that. Actually, the problem I have is that I only get 15-20 different samples per second from my controller (when PPSSPP asks more frequently, it just gets repeats.) It seemed like I had about a 0.1s latency or so (which is way high in my book), but that seemed like a hardware problem for me.

I don't even think there were any changes in the DirectInput or XInput code since v0.9.6.

So you're really the person best able to investigate it. If you try 10 builds or so, you should be able to narrow it down to exactly where it started for you. Just go by halves, it's the mathematically most efficient way. Statistically speaking, if you can narrow it down like that it's likely it'll get fixed, since we can see what change caused it.

-[Unknown]

[tsukiyomaru0]

I have a suspicion... Has there been any change in the development platform? I do recall that, after installing the new version, I had to get a new version of the NET library before using... And is really, just delay from gamepad input that increased that much, and I have been using the same port before and now.

And I can't find the old builds between 0.9.6 and 0.9.7.

[Henrik]

Lots of builds between 0.9.6 and 0.9.7 are available here:

http://buildbot.orphis.net/ppsspp/

[tsukiyomaru0]

(03-09-2014 06:18 PM)Henrik Wrote:  Lots of builds between 0.9.6 and 0.9.7 are available here:

http://buildbot.orphis.net/ppsspp/

Either way, tracking it this way will take weeks, months tops. There was something regarding input/output that was changed between them, as well as something regarding the libraries required to run it as, I mentioned and repeat, it inquired me to install an updated NET Library.

[solarmystic]

(03-11-2014 11:43 PM)tsukiyomaru0 Wrote:  Either way, tracking it this way will take weeks, months tops. There was something regarding input/output that was changed between them, as well as something regarding the libraries required to run it as, I mentioned and repeat, it inquired me to install an updated NET Library.

Not exactly. With the method of bisecting, you'll be able to track it down really quickly. Here's a rather helpful post in the ppsspp github wiki written by [Unknown] himself to get you started.

In any case, the devs will not be able to help you out unless you provide them with the exact build that's responsible for the issue that you're experiencing. Help them help you solve the issue.

[[Unknown]]

Also, we do not use .NET. However, we are using a newer Microsoft Visual C++ runtime. If this is somehow causing problems for your controller driver, it may be a bug or poor interaction with your specific controller driver. Potentially updating it may help.

As far as bisecting: even without git, you can go by halves. It's the most efficient way when there's a point-in-time to find. The common example is "guess a number, 1 to 100", "50", "too high", "25", "too low", "35", etc. It's provable (usually in introductory computer science courses) that this is the fastest way to find the number they guessed, by going in halves.

From 1-1000 would take at most 10 guesses to get exactly. Try it if you don't believe me.

Even if you don't get it exactly, a range of say 10 commits would be good enough probably (exactly would be a lot better.) So if you can try ~10 builds you can definitely narrow it down pretty quickly.

-[Unknown]