Module kernel_args 

Type-safe kernel argument builder for CUDA + Metal kernel launches.

Replaces manual Vec<*mut c_void> construction with a builder pattern that prevents parameter type/order mismatches AND records per-arg type information so the metal backend can dispatch buffer args via setBuffer:offset:atIndex: and scalar args via setBytes:length:atIndex: (cuda’s untyped cuLaunchKernel cannot distinguish the two; metal cannot conflate them).

Usage

KernelLaunch::new(gpu, kernel)
    .grid([num_tokens, 1, 1])
    .block([256, 1, 1])
    .arg_ptr(input)
    .arg_u32(hidden_size)
    .arg_f32(eps)
    .launch(stream)?;

Internally, every arg is recorded with its kind (Buffer / Scalar) and its native byte width. launch() materializes a typed KernelArg slice and calls GpuBackend::launch_typed. The cuda backend’s default launch_typed impl flattens that back into the legacy void** shape; the metal backend overrides launch_typed to thread the type info through to the encoder.

Structs

KernelLaunch

Builder for type-safe kernel launches across CUDA + Metal.

Functions

div_ceil

Convenience: divide and round up.