Add support to invoke HiFi SIMD kernel for FP32XFP32 SUB Operator

[anupampandey-ittiam]

The HiFi5-optimized SIMD kernel for the FP32XFP32 SUB operator is available in the xa_nnlib_hifi5 library. However, there is currently no integration or framework-level support to invoke this optimized (i.e., HiFi5 SIMD) implementation in TFLM.

Current Behavior:
Invokes ‘RefC’ implementation of FP32xFP32 SUB operator while running inference on Xtensa with optimizations enabled

Expected Behavior:
Should invoke available HiFi5 SIMD optimized FP32XFP32 SUB operator while running inference on Xtensa thereby improving performance

Proposed Fix:
Add the support in TFLM to invoke the available HiFi5 SIMD of FP32XFP32 SUB operator