New LLE RSP Plugin (accurate, fast but untested?)

[zilmar]

I have been going through the pull request, in general it looks all good, and I probably should just accept it, just doing more testing and looking at the code, just do not have a lot of time.

[HatCat]

Thanks for the information V1del.

zilmar you might be tolerant of it, and I can't oppose that, but hopefully I'm saving both him and you time in the future pull requests to come so it doesn't take you so long in later ones.

[Melchior]

HUWayyy Zilmar THE GRAND, HE LIVES

[HatCat]

went ahead and committed that CreateFile thing rpg pointed out.

Doesn't affect anything but I figured it out, may as well get it over with in case it matters later.

[RPGMaster]

Good to hear !

Something's been confusing me for a while. I've been reviewing LWC2/SWC2 instructions lately, and I noticed something unusual about your SDV edge case implementation.

for (i = 0; i < 8; i++)
            DMEM[BES(addr &= 0x00000FFF)] = VR_B(vt, (e+i)&0xF);

Is this some kind of mistake? addr never changes, besides maybe the first time. I noticed this piece of code is the reason WDC with Jabo 1.7 looks different with your RSP than other plugins.

[HatCat]

I know about that. It's a brain fart I couldn't help but notice while re-reading the source.

I didn't fix it--I don't know why. Maybe I wanted to wait for somebody to see where a game used it.

https://github.com/cxd4/rsp/commit/c...c473fb34dL1184
That was the commit that broke it. I had a brain fart when optimizing SDV--the LS_Group_I function was my implementation of the documentation SGI's patent had on it, which was definitely bug-free (but like 15 times slower).

I don't really want to know how it looked on D3D8 though...could you compare screenshots in angrylion's plugin of how the game looks?

[HatCat]

Dammit, and stop quoting my code with weird spacing!

you still put stuff like this

for (i = 0; i < 8; i++)
            DMEM[BES(addr &= 0x00000FFF)] = VR_B(vt, (e+i)&0xF);

Don't cause people to think I space my shit like that dammit. At the very least use tabs, instead of 12 spaces.

[RPGMaster]

Rofl, I have a bad habit of copy pasting, with the first line of code not being tabbed. I almost forgot that you already implemented a way to play frame by frame. I assumed you wanted filters off.
The left image = before fixing. The right = after.


For the change, I just did

for (i = 0; i < 8; i++)
    DMEM[BES(addr + i) & 0xFFF] = VR_B(vt, (e + i) & 0xF);

This is correct right? Just want to be 100% sure .

Idk if you're interested in optimizing edge cases, but for SDV & SQV edge case, I've only seen e = 12. I'll probably start testing more games after I finish refining algorithms, now that I'm curious about certain things.

V1del, thanks for the info.

[blikblum]

Originally Posted by RPGMaster

Is there a way for me to choose which commits to submit in a pull request? Otherwise, I'll generally just wait until it gets accepted, before pushing new commits.
.

There are some options, here's one:

Create a new branch from the last commit that you want to be in the Pull Request (PR):

git branch name-of-the-branch <sha1-of-commit>

Push to github

Go to github homepage, select the new created branch and create a new PR.

Only changes made to this branch will get in the PR. If you want to modify something in that branch do a checkout and make commits

git checkout name-of-the-branch

[do commits]

to get back to master

git checkout master

It's advisably always to create a branch for each fix / feature, so makes easy to contribute back and continue working

[HatCat]

Thanks blikblum for extending on the Git usage explanations and options for us.

Originally Posted by RPGMaster

Idk if you're interested in optimizing edge cases, but for SDV & SQV edge case, I've only seen e = 12.

Mmm...optimize too many edge cases and the interpreter starts to look like a recompiler.

I only did a switch on the addr/element alignment in SWC2 because unaligned instructions are illegal--which like any other invalid/illegal RSP operation has to execute anyway without exceptions. Still, for documentation sake (and for much better accuracy and performance), I only singled out the address' 64-bit alignment (or lack thereof) for optimizing parallel data moves. This fails only for games like World Driver Championship which of course uses unaligned, illegal SDV instruction.

Besides, even if I were to optimize exact game-specific cases, it would be less like an interpreter, more like a recompiler. Too many hard branches combining element/address alignment and it's hardly any different than working on a recompiler or HLE game hacks really.