Build-time PTX compilation for warp-types GPU kernels.

Use in your build.rs to cross-compile a kernel crate to PTX, then load the generated PTX at runtime via cudarc.

Example

// build.rs
warp_types_builder::WarpBuilder::new("my-kernels")
    .build()
    .expect("Failed to compile GPU kernels");

Then in your main crate:

// src/main.rs
mod kernels {
    include!(concat!(env!("OUT_DIR"), "/kernels.rs"));
}

fn main() {
    let ctx = cudarc::driver::CudaContext::new(0).unwrap();
    let k = kernels::Kernels::load(&ctx).unwrap();
    // k.butterfly_reduce — CudaFunction handle ready for launch
}