In my local copy, I have reverted 1e00486 and the above test case functions correctly.

I'll assign this issue to scztt to review this commit.

Before my change, the SIMD cases of LinXFade were reversed, so reverting won't do anything but trade one bug for another.

I believe the (1.0f - ...) matches the amp that's required in LinXFade2_next_k_nova - but the reversal isn't appropriate for other places that use unit->m_amp. I think:

1. unit->m_amp should be reverted to (unit->m_pos * 0.5f + 0.5f), so that m_amp corresponds to the amp of the second (right) channel.
2. LinXFade2_next_k_nova and any other LinXFade2 functions should be swept to make sure they're using this definition of unit->m_amp and unit->m_pos.

I think it might be just LinXFade2_next_k_nova that needs to be changed, but it would be good to check at least? Can you send me unit tests for the cases that you're seeing failures, so I can make sure I get those cases to pass?

I think you're right about the problem being specifically the nova functions.

LinXFade2_Ctor shows three cases, based on the rate of the pan input:

• Scalar rate: LinXFade2_next_i_nova
• Control rate: LinXFade2_next_k_nova
• Audio rate: LinXFade2_next_a

Audio rate is the only one that doesn't use nova, and in the following test, it's the only one that works correctly.

a = {
    var a = DC.kr(0), b = DC.kr(1), iTrig = Trig1.kr(Impulse.kr(0), 1),
    pans = [-1, DC.kr(-1), DC.ar(-1)];
    Poll.kr(iTrig, LinXFade2.kr(a, b, pans));
    FreeSelf.kr(iTrig <= 0);
    Silent.ar(1)
}.play;

UGen(LinXFade2): 1  // calc. by LinXFade2_next_i_nova -- wrong
UGen(LinXFade2): 1  // calc. by LinXFade2_next_k_nova -- wrong
UGen(LinXFade2): 0  // calc. by LinXFade2_next_a -- correct

I'm too short of time today to rebuild SC without nova optimization, but the formulas in the nova-less _k and _i functions match _a, so I believe they are also okay.

I agree with your proposed solution: unit->m_amp should follow the user's input value (for clarity if for no other reason). Since the flaw is really in the _nova functions, then these should be fixed.