rtlamr seems single threaded

[kevarh]

rtlamr seems single threaded

[bemasher]

The main problem with building concurrency into rtlamr is that each stage depends on the result of the previous, the best we can do is pipe-lining. This doesn't necessarily help anyway because each block of data must be processed entirely by one section of code before it can be handed to the next.

Splitting the blocks into smaller pieces would increase granularity of execution and allow each section to spend more time processing and less time waiting but this will likely sacrifice throughput because most of the DSP related tasks are optimized to the point that main memory accesses are the bottleneck in most systems rather than the CPU.

I'll do some testing to see if this will help any on lower power systems, I'm betting the only place we can add concurrency would be splitting Pack and Search off from the rest of the execution.

[bemasher]

I've done some investigation into this and there isn't much opportunity for performance improvement by separating portions of the signal chain into goroutines. There will be some benefit to this when multiple protocols per invocation is supported but until then I'm closing this issue.