Before I jump into the complicated world of interfacing with C and C++, I'll introduce a simpler example—a library written in Delphi. The motivation for its use comes not from a bad algorithm that Mr. Smith wrote, but from a badly performing 64-bit compiler. This is not something that I am claiming without proof. Multiple Delphi programmers have pointed out that the 64-bit Windows compiler (dcc64) generates pretty bad floating point code that is 2-3 times slower than the floating point generated from an equivalent source by a C compiler.
When you have already explored all the standard approaches of speeding up the program, and the compiler is the only source of the problem, you cannot do much. You can only rewrite parts of the program in the assembler, or use an external library that works faster than the native code. Such a library will either use...