David Chocholatý

> Thanks for adding benchmarks. A couple of thoughts: > > Irregardless of the buffer size, the relative performance stays the same. Also, since we're benchmarking realtime performance, the standard...

> > Thank you for the nice result table from your system. > > I forgot to say that those actually came from CI, so the absolute values might not...

> IMO you can leave them as is for now. I just wanted to share some general thoughts on how we currently do our microbenchmarks. If you want to look...

> Another inspiration for a singlethreaded Ringbuffer: https://hg.sr.ht/~breakfastquay/rubberband/browse/src/common/SingleThreadRingBuffer.h?rev=tip Thank you for this tip.

I would like to post a little info about the possible choice of using only `memcpy` for `RingDelayBuffer` or `SampleUtil::copy` instead (vectorized loop for SSE on 32-bit). Because my system...

I think, that I should just summarize some of your reviews and the current code look. I would like to discuss, the problem, for most of the review's indirectly points....

> Irregardless of this, I still don't understand why the read position calculation has to be as complicated as it currently is. The read position calculation can be of course...

> I don't understand how it can happen that the read position is ahead of the write position. That would mean that you are trying to read more samples than...

Okay, perfect. Thank you. So now when we agreed to allow the mentioned special case, I will simplify all the calculations asap.

> When it happens that you read out the initial zeros, you will hear a click sound when the real samples starts. This can be avoided by fading in the...