Alcides Fonseca

In the compilers course that I teach at the University of Lisbon students have to implement a compiler for a given language, and generate LLVM IR. While students have the freedom to choose any language they wish for the compiler, Java is a popular choice due to their familiarity with the language.

One particular issue students had was to choose the right constant for a given floating-point number.

int main()
{
    float a = 3.2;
    float b = 3.9;
    return 0;
}

is compiled to

define i32 @main() #0 {
  %1 = alloca i32, align 4
  %a = alloca float, align 4
  %b = alloca float, align 4
  store i32 0, i32* %1, align 4
  store float 0x40099999A0000000, float* %a, align 4
  store float 0x400F333340000000, float* %b, align 4
  ret i32 0
}

Now the goal is to generate “0×40099999A0000000” from 3.2, “0×400F333340000000” from 3.9 and so on.

To solve this issue, I extracted and refactored code from Scala-Native, which you can find below.