Our code uses the ceres library. The whole project runs on C7, which involves floating point matrix operations. We want to use mma to speed up the matrix operation process. We find that we can only do integer matrix operations, but not floating point matrix operations. We can compile successfully, but the operation is stuck. What methods can we use MMA?
At the same time, I found that the matrix budget of dsplib can support floating point and is fast, but dsplib can only run on C6, while our code needs to run on C7. I plan to make some plans. Can you give me some suggestions？
Scheme 1: Compile the code running on C7 to C6, but the running speed of the whole project is extremely slow.
Scheme 2: Compile the source code of dsplib to the C7 platform. Is this scheme OK?
Scheme 3: Open a C6 core to run the matrix budget separately, use ipc communication, wait for the data sent from C7 to be processed by the C6 DSPLIB operation, and then send the results to C7. What is the cost performance of this scheme, and will the communication time consumption be long?
