C++ Optimization of Convolution using NEON and ASM

Job Description: We have a C++ Code that runs on a Boundary [login to view URL] Development Board. We have implemented an optimized convolution for NEON-ASM with 32 bit floats, but we seek to further decrease the execution time using 8-bit-integer calculations on the NEON-Unit instead. We provide as a starting point hardly optimized reference implementation with 8-bit integer arithmetic in C++ which produces the desired output and generates runtime measurements. We also provide optimized NEON-ASM implementation with 32 bit floating point calculations providing the reference execution time which should be beaten by a factor 2. Your task would be to optimize the 8-bit integer run time using C++, NEON and ASM optimization and implementation techniques. We expect a speed up of factor 2 in comparison with the 32 bit NEON ASM floating point implementation. We will provide ssh access to a [login to view URL] boundary development board with the gcc-compiler-toolchain for testing and measurements. The code fragment will be on that board. Also the provided reference output should not change significantly when the optimized method is executed. Constraints: 1. You must be fluent in English or German. 2. You must be familiar with programming NEON and ASM. 3. You should have experience with optimizing calculations for ARM 4. Please deliver your results with measurements, and a short description.