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 struct
LamportGeneration
struct LamportGeneration
Owns the three-buffer generation rotation for a Lamport exchange.
The classic Lamport hazard is a fast rank racing into call N+1 and overwriting a slot a slow rank has not yet read for call N. Three rotating buffer generations defend against it: each call writes-and-reads one generation while clearing the generation that will be reused two calls later, leaving exactly one generation of skew slack.
Given a monotonically-increasing call counter flag:
data_index=flag % 3is the generation written and read this call;clear_index=(flag + 2) % 3is the generation cleared back to the sentinel this call (it will next be written two calls from now).
data_index != clear_index always holds, which is the one-generation-of-
slack invariant. This mirrors circular_add's pure-integer style in
sync.mojo so the kernel body and any future broadcast consumer share
identical semantics.
Implemented traits
comptime members
NUM_GENERATIONS
comptime NUM_GENERATIONS = 3
Number of rotating buffer generations: read-this-call + write-next-call + clear-for-the-call-after.
Methods
data_index
static def data_index(flag: Int) -> Int
Returns the generation index written and read on this call.
Args:
- flag (
Int): The monotonically-increasing call counter.
Returns:
Int: flag % NUM_GENERATIONS.
clear_index
static def clear_index(flag: Int) -> Int
Returns the generation index cleared to the sentinel on this call.
This is the generation that will next be written two calls from now; clearing it now overlaps with the current exchange.
Args:
- flag (
Int): The monotonically-increasing call counter.
Returns:
Int: (flag + NUM_GENERATIONS - 1) % NUM_GENERATIONS, i.e. (flag + 2) % 3 for the default of three generations.
Was this page helpful?
Thank you! We'll create more content like this.
Thank you for helping us improve!