For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /get-started.md).

Mojo function

gmem_edge_masked_load

def gmem_edge_masked_load[dtype: DType, src_space: AddressSpace, //, width: Int](ptr: Pointer[Scalar[dtype], address_space=src_space], mask: Int16) -> SIMD[dtype, width]

Loads an edge-masked vector, treating ptr as device (global) memory (Apple M5 only).

Like edge_masked_load but reinterprets any pointer as global — for a generic-typed pointer to device memory. Emits AGX3 load.with.emask.global.

Constraints:

width in {1, 2, 4}, dtype 8/16/32/64-bit. Apple M5 only.

Parameters:

• ​dtype (DType): Element type of the pointer (inferred).
• ​src_space (AddressSpace): Pointer source address space (inferred).
• ​width (Int): Number of elements to load.

Args:

• ​ptr (Pointer[Scalar[dtype], address_space=src_space]): Base pointer of the load, pointing to global memory.
• ​mask (Int16): Per-lane active mask, e.g. from build_edge_mask.

Returns:

SIMD[dtype, width]: A SIMD[dtype, width]; in-range lanes loaded, masked-off lanes zero.