Which parts of CMSIS-DSP are used for CMSIS-NN? I'm asking because I want to disable as much options as possible of the CMSIS-DSP.
CMSIS-NN is a lib accelerating neural network kernels that can help TFLM and TVM accelerate inference while CMSIS-DSP is a lib for more mathematical calculations. ARM_MATH_MVEI and ARM_MATH_DSP flags are defined automatically (in arm_math_types.h) depending on the CPU for which you compile and based upon compiler #defines. But generally, both are defined since your core kernels can have a vector part using MVEI extensions and a scalar part using DSP extension. You can disable either using these flag options.
The functions supporting TensorFlow Lite framework is identified by the _s8 suffix and can be invoked from TFL micro. The functions are bit exact to TensorFlow Lite. Refer to the TensorFlow's documentation in [3] on how to run a TensorFlow Lite model using optimized CMSIS-NN kernels.