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

Mojo function

store_cluster_smem

def store_cluster_smem[dtype: DType, width: Int](local_ptr: Pointer[Scalar[dtype], address_space=AddressSpace.SHARED, _safe=False], peer_rank: UInt32, val: SIMD[dtype, width])

Store val into peer peer_rank's shared memory at local_ptr.

Symmetric to load_cluster_smem: writes the width elements into the same shared object as it exists in CTA peer_rank. Bracket cross-CTA writes with cluster_sync() so the peer observes them. 32-bit element dtypes only.

Parameters:

dtype (DType): Element dtype of the shared buffer; must be a 32-bit dtype (inferred).
width (Int): Number of elements in val to store (inferred).

Args:

local_ptr (Pointer[Scalar[dtype], address_space=AddressSpace.SHARED, _safe=False]): Pointer into this CTA's shared memory identifying the shared object to write.
peer_rank (UInt32): Target CTA rank whose copy of the shared object receives the write.
val (SIMD[dtype, width]): Vector of width elements to store into the peer's shared memory.