Detect vector+! and a few other common inlined vector ops

[water111]

There's also one with a multiply in there.

• vector-!
• vector+!
• vector-reset!
• vector-float*!
• vector-dot 3
• vector-dot 4

[water111]

All but vector-float*! are implemented in #637. But I'm starting to think that there's a better way for some of these that detects at the instruction level.

    lqc2 vf1, 0(gC)
    mfc1 gD, fA
    qmtc2.i vf2, gD
    vaddx.w vf1, vf0, vf0
    vmulx.xyz vf1, vf1, vf2
    sqc2 vf1, 0(gE)

where:

fA = float in
gC = vector in
gD = temp
gE = vector out

It will make propagating in the float much easier in vector-float*!, and I think will reduce the chances of a bug in the propagation logic (currently it might be possible for the arguments to vector+! to be evaluated in the wrong order).

vector-reset! should stay as it is though.

[water111]

Now vector+!, vector-!, vector-float*!, and vector-dot! (3) all are recognized at the instruction level and it works well.

It looks like vector-cross is sometimes inlined, but not always. We should find if this is actually another function being inlined or not, and support it in a similar way.

I haven't seen any examples of vector-dot 4 yet, so I'm not going to worry about that for now.

[water111]

vector dot 4 is in baseplat. https://github.com/xTVaser/jak-project/tree/decomp/plat

[water111]

vector 3 is not working in camera.gc, likely because of stack variables.

[water111]

vector4-dot added in #803.
I think supporting vector ops which involve stack spills is not worth the effort. We detect dot products by looking at instructions and it would be a lot of work to move this to later.

The others have been working extremely well, no issues so far.