Enum KernelArg 

pub enum KernelArg<'a> {
    Buffer(DevicePtr),
    Bytes(&'a [u8]),
}

Typed kernel argument, used by launch_typed.

CUDA’s cuLaunchKernel is type-blind — every arg is void* and the driver interprets bytes by kernel signature. Metal’s MTLComputeCommandEncoder is not: buffer arguments require setBuffer:offset:atIndex: (the encoder tracks the resource) while scalar/struct args require setBytes:length:atIndex:. KernelArg preserves that distinction so both backends can dispatch correctly.

Variants

Buffer(DevicePtr)

A device buffer at this base GPU address. The metal backend resolves it to its owning MTLBuffer + offset via the alloc registry; the cuda backend forwards the raw u64 to the driver.

Bytes(&'a [u8])

Inline scalar/struct bytes, e.g. a u32 count or an f32 eps. Length is forwarded to Metal’s setBytes:length:; the cuda backend zero-pads up to 8 bytes per slot.

Trait Implementations

impl<'a> Clone for KernelArg<'a>

fn clone(&self) -> KernelArg<'a>

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl<'a> Debug for KernelArg<'a>

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl<'a> Copy for KernelArg<'a>