Use TVM for TI DSP


Hello geeks, first time here.
I am learning TVM now and I am working on a TI DSP with two c66x cores.I have seen that TVM can generate souce code for CUDA/OpenCL, so I am wondering if I could extend this to generate optimized c66x souce code, which use “#Pragma roll/iterate before the for loop” and “restrict keyword before the variable” to accelerate the loops. Would you give me some suggestions?


You can start by looking the codegen for some backends, such as opencl (already uses _restrict), HLS (uses pragma).


Thank you for your reply.

I will try to implement it.