This is a set of simple benchmarks to estimate a baseline of the cost of common scenarions when C# works with native code through DllImport & friends.

The test code runs a large number of operations and averages out over the number of operations, as the resolution of Stopwatch is not small enough.

• Mono >= 5.0
• Dotnet >= 3.0, with both netcoreapp3.0 and netcoreapp2.2 frameworks.
• python with pandas and seaborn for plotting the graphics

$ make
$ make plot

These plots discard the initial iterations as they happen to be quite slow due to the initial JIT'ing of the managed code. The x axis shows the number of iterations and the y axis shows the average time per iteration in microseconds.

• ManagedStatic - Simple call to an empty static method.

• ManagedVirtual - Simple call to a virtual instance method

• DllImportDirect - Call to fprintf redirect to /dev/null

• DllImportCustomMarshaller - Call to fprintf but using an empty ICustomMarshaler for the string being passed.

• DllImportWithCallback - Call to native qsort passing a C# callback.

• Dell XPS 15 9560
• i7-7700HQ @ 2.8 GHz
• 32 GB RAM
• Ubuntu 19.10
• Mono JIT/mcs 6.4.0.198 from Mono's repo.
• dotnet 3.0.100 from Microsoft's repo, targeting netcoreapp both 2.2 and 3.0