It has a C compiler, so you should be able to use shared memory and run a separate thread on the DSP to do whatever you need. Since it's a DSP, it has a dual multiply and accumulate capability among other special instructions. So, it's crazy fast, but you have to have a low level view of the chip to make use of it well.
The main reason many people shy away from the NEON stuff is that with the A8 the NEON's not fully pipelined in with the regular instructions. You need to learn a bit of finesse with their use or you can cause an instruction pipeline stall instead of a performance boost.
Originally Posted by gbraad
And it's a bit limited in what it can/can't do. The best use of the DSP is doing video decode or compressed audio decode or similar classes of tasks, including the compression direction of that pathway.
Originally Posted by garytr24