Performance Hit When Using Raphnetraw v1.0.7

[KojiKenshiro]

Let me say that I love the prospect of using OEM controllers & peripherals with Project64, but there's a noticable performance hit while using the pj64raphnetraw_1player.dll plugin with a 2-Port N64 USB Adapter. Any insights into a possible fix that could rectify said issue would be greatly appreciated.

Edit: I can now confirm that inserting a Rumble Pak (or any other Accessory?) causes Project64 to have some performance issues. When using just the Controller, the performance is as it should be. Would allowing the Plugin to run on a separate thread from the emulator possibly fix this issue?

Thank you for taking the time to allow us the ability to use OEM controllers & peripherals with Project64.

[PFLI2]

I believe we're having the same issue. Seems 3 of the 5 issues open are related:
#8
#3

[Ecksters]

In my original issue, it was discussed that if it were a USB speed issue there likely wasn't much that could be done, but given you're indicating that a faster PC seemed to mitigate the issue somewhat, I wonder if that indicates it's actually a performance problem outside of the hardware.

[PFLI2]

In my original issue, it was discussed that if it were a USB speed issue there likely wasn't much that could be done, but given you're indicating that a faster PC seemed to mitigate the issue somewhat, I wonder if that indicates it's actually a performance problem outside of the hardware.

I thought the same thing before I built an entirely new PC and found the same issue only slightly improved. I'm positive this is a software thing at this point.

[raphnet]

I'm sorry, I want to look into this but other stuff keeps coming up all the time and I never seem to get to it... It appears that the more or less 1ms round-trip to exchange data with each controller add up (especially with rumble paks, which implies two exchanges) and this blocks the emulator.

So what I would like to try is a new asynchronous plugin architecture where the plugin emulates a N64 controller (so it can provide an immediate response to the emulator) while in background continuously polling the real N64 controller while also forwarding requests for rumble. Support for controller paks (and other devices) would still need to happen synchronously (although a caching layer for mempak could be added), but having the controller polling happen in background instead of blocking the emulator should already help a lot.

Now it's "just" a matter of finding the time to implement and test the above...