This file proposes some benchmark performance in both C++, Python and Julia. The purpose is to estimate the output throughput both when a Software Defined Radio (SDR) and in direct timing.

Hardware configuration

The following configuration has been used

• i7-8850H CPU @2.60GHz with Intel SSE4.2, Intel AVX2
• 32GiB System Memory (4* 8GiB SODIMM DDR4 2667 MHz)
• Ettus X310 with UHD 3.15 driver
• FFT realized with FFTW 3.3.8 (full compile flag), plan generated in patient mode for all language

Language configuration

We expose here the specific properties uses for each langage.

C++

• Flag -gnu++11 -O3 -march=native -mavx
• Gcc 7.5.0

Julia

• Flag -O3
• Julia 1.5.0-rc1.0
• LoopVectorization (0.8.8)
• StaticArrays (0.12.3)
• FFTW 1.2.4

Python

• Flag -OO
• Python 3.6.9
• numba (0.50.1)
• llvmlite (0.33.0)

To compile and launch benchmark

We expose here the steps to compile the codes and obtain the performance. There is 2 main folders

• Receive: Different code to be used with a SDR (i.e X310) to monitor link between X310 rate and processing rate
• SoftThroughput : Maximal rate in different langage with no interface with SDR with 3 optimisations level

C++

To compile the file and run the bench use the follwing command: build && cmake .. && make && ./benchCpp

Julia

In Julia, the julia benchmark files can be launched in a fresh REPL or alternatively by julia -O3 benchmark_rate_severalRate.jl

Python

The file to be launched in Python is python3 -OO benchPy.py