Best practices

Best practices

The program should work as double precision (or custom precision). There is no place to give a choice to the compiler. When you write code you must know what precision it should be, that is why we need to use the following rules:

• no default declarations of real. Use parameter for:

  • single precision real(kind=singleprecision)
  • double/custom precision real(kind=prec)

• function real should contain precision

  • single precision real(variable/formula, kind=singleprecision)
  • double/custom precision real(variable/formula, kind=prec)

• numerical values should contain information about precision

  • single precision e.g. 1.0_singleprecision 0.0_singleprecision 1.e-12_singleprecision
  • double/custom precision e.g. 1.0_prec 0.0_prec 1.e-12_prec

• if you need use different precision than used in program add it in size.f90 file and use it as previous examples

• math functions sin(), cos(), asin() etc. are in default precision. Use compiler flags for single or double precision as default

  • gfortran -fdefault-real-8

• ifort -r8
• pgf90 -r8

• in md.f90 are defined one and zero use them in that file:
  • real(kind=prec) :: one = 1.0_prec
  • real(kind=prec) :: zero = 0.0_prec

Every time you don't use best practices you are changing the resulting values for default precision. We do not want this.