libthermo

A fast C++ thermodynamical gas library.

libthermo targets high-performance computing using really simple/simplistic thermodynamical modelings instead of complex and more accurate approaches (such as https://cantera.org/).

It's well fitted for industrial simulation where the base blocks need to be efficient to not become bottlenecks.

Modelings

Properties

The library provides modelings for the following properties:

Properties

For now, only ideal gas is implemented in IdealGas class.

It has constant specific heat pressure, thus ratio.

Benchmarks

Benchmarks can be ran using:

• C++: thermo_benchs CMake target
• Python: ipython python_benchs.ipy

The metrics are computed over 1M elements for operation only (allocation is not timed), and are given in nanoseconds/element:

IdealGas

*pythermo = Python bindings, incl. xsimd and tbb (vectorization and multithreading)

Conclusions are:

• Pure NumPy (vectorized), C++ loop over a 1D buffer allocated on the heap (std::vector) and xtensor are directly comparable
  • xtensor shows performance issues on very simples cases (Cp, gamma, H): to be investigated
• xsimd and tbb, turned on by simply adding 2 CMake flags give a huge benefit
  • up to x1.65 speed-up for xsimd
  • up to x15 speed-up for tbb, on a 12-cores/24-threads AMD Ryzen 3900XT
• using xtensor-python Python bindings to operate on NumPy buffers gives the same performance as pure C++ code