Vocoder plugin: refactor for speed

[softrabbit]

Changing from arrays of structs to structs of arrays makes the code a lot more optimizer-friendly, with possible speed gains of 2-3x when compiling for SSE systems.

(see https://github.com/softrabbit/filterbench/tree/master/vocoder for some benchmarking code)

[Wallacoloo]

Wow, those are some significant performance gains!

I ran your benchmarks on my 64-bit Athlon II P320 (mobile processor from 2009 with SSE, SSE2, SSE3 and AMD's SSE4a):
gcc 4.9.1:

10.73 ./original
10.74 ./sse
5.38 ./optimized

clang 3.6.0 (modified your Makefile & removed the incompatible -fopt-info-vec-optimized flag) :

10.16 ./original
10.14 ./sse
5.59 ./optimized

So the improvements seem to generalize pretty well. And the outputs of the new version match the old for both gcc and clang builds.

Nicely done!

[curlymorphic]

nice work @softrabbit , have you issued a pull request upstream?

[softrabbit]

nice work @softrabbit , have you issued a pull request upstream?

Nope, not yet. AFAIK the version in LMMS is adapted a bit from the upstream version, so it's probably not a straight pull. And then there's the standalone JACK version, that might need similar treatment.