[PanDaMan]

@PanDaMan - DaniellaE's branch will also be clipping, this is an existing issue, what's different is that AE reports it so we know when it occurs. We are currently working on a DRC patch that will help with this behaviour instead of just clamping the buffer. (We never output clipped audio, we have a soft clamper that performs a poor mans DRC to the samples when this occurs).

@gnif - Thanks. I did not hear any distortion with DaniellaE's, I also don't hear anything or get any clamping logged if I set the volume in guisettings to 0.999999 which should be close enough to 1.0 not to matter. See http://xbmclogs.com/show.php?id=2103
My C is very rusty, but wouldn't the missing 'f' suffix on line 939 in SoftAE.cpp mean it's being compared as a double rather than float and could the promotion be marginal making it think it needed to clamp when it didn't?

By "poor man's DRC" did you mean the tanh approximation? That looks well off for values |x| > 0.5 if I'm reading the plot here correctly: http://bit.ly/Jx4ULG Applying this only on blocks with values > 1.0 means there's additional non-linearity between blocks that get clamped and those that don't. That might also cause the distortion I was hearing.

Finally, the SSE clamp looks broken, implementing (x * (c1 + x^2)) / (c2 + x^2). c2 should just be 9.0f and an additional multiply operation is needed (x * (c1 + x^2)) / (c1 + (c2 * x^2))