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

calculate_warp_offset

def calculate_warp_offset[MaskType: DType](state: Bool) -> Tuple[UInt64, UInt64]

Computes warp-level write counts and per-thread offsets using warp voting.

Given a per-thread boolean vote, returns the total number of threads in the warp that voted true and this thread's write offset among the preceding threads that voted true.

Parameters:

MaskType (DType): Unsigned integer DType wide enough to hold one bit per thread in the warp.

Args:

state (Bool): Per-thread boolean indicating whether this thread contributes a write.

Returns:

Tuple[UInt64, UInt64]: A tuple of (writes, offset) where writes is the total number of threads in the warp that voted true, and offset is the number of preceding threads (lower thread IDs) that voted true.