Integral and differential calculus is not useful at all [1]. Compilers are not very computation-heavy, and where they do require grinding through math, the interesting problems are discrete math rather than continuous variable problems, which means that integral and differential calculus very rarely offer insights.

The math you usually want is algebra. Even then, the elementary algebra you learned in high school (or earlier) is usually sufficient. More complex domains in math are usually used for theoretical analysis and are less common in practice. Many dataflow optimizations can be described as monotonic functions over a semilattice, but they are rarely implemented as such. Type systems in functional languages can be understood more richly as applied category theory. Linear algebra is useful when multidimensional loop nests come into play.

[1] Unless maybe you're building a neural network compiler.