notes

Author: catfact

notes.md

@@ -0,0 +1,45 @@
+idea for new voice computation structure,
+
+should allow "duck" and "follow" featurees without compromising performance.
+implementation complexity is reasonable;
+the cost is some extra RAM usage (perhaps several KB per voice.)
+
+1. allocate a number of buffers at least as big as the largest expected blocksize. each buffer will contain the value of a parameter for a whole block:
+- active subhead index
+- each subhead state:
+  - read phase
+  - write phase
+  - pre+fade level
+  - rec+fade level
+
+2. each process block can be decomposed without looping over all voices on each sample. 
+
+## normal mode:
+
+same as present operation, with some potential speed gain due to memory locality:
+
+- update positions
+- update fade levels
+- peek+poke, mix
+
+## "follow" mode:
+
+here, one voice is always hard-synced to another, but with different buffer data and mix parameters. this allows stereo operation among other things.
+
+- copy positions
+- copy fade levels
+- peek+poke, mix
+
+## duck mode:
+
+one voice uses another's write index as a reference; output level is attenuated when this voice's read position crosses that voice's record position.
+
+- update positions
+- update fade levels, comparing with reference position
+- peek+poke, mix
+
+
+### details
+
+- can use `std::array` and `std::copy` for state buffers
+-