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 struct
BundledBroadcast
struct BundledBroadcast
Registers the mo.bundled.broadcast graph op with the graph compiler.
The per-device form of mo.distributed.broadcast: one instance per GPU
copies the root device's tensor into its own output. execute reads the
root tensor through its live data pointer; build_graph reads it through
a stable address slot (a Stable input), so the recorded device graph
replays against whatever buffer the slot names at the time.
Implemented traits
Methods
execute
static def execute[Sigs: TypeList[Sigs.values], //, root: Int, _trace_name: StaticString, peer_rank: Optional[Int] = None](output: T, input: T, signal_buffers: TensorTuple[*Sigs.values], ctx: DeviceContext)
Per-device broadcast through live addresses.
Parameters:
- Sigs (
TypeList[values]): The element types ofsignal_buffers(inferred). - root (
Int): Index of the root device in the signal buffer list. - _trace_name (
StaticString): Trace name used for profiling and debugging. - peer_rank (
Optional[Int]): THIS GPU's index in the signal buffer list;Nonetakes it from the device id.
Args:
- output (
T): Output tensor for THIS GPU. - input (
T): The root GPU's tensor. - signal_buffers (
TensorTuple[*Sigs.values]): Signal buffers for ALL participating GPUs. - ctx (
DeviceContext): Device context for THIS GPU.
build_graph
static def build_graph[Sigs: TypeList[Sigs.values], //, root: Int, _trace_name: StaticString, peer_rank: Optional[Int] = None](output: T, input: T, signal_buffers: TensorTuple[*Sigs.values], builder: LegacyDeviceGraphBuilder)
Records the per-device broadcast into a device graph.
Parameters:
- Sigs (
TypeList[values]): The element types ofsignal_buffers(inferred). - root (
Int): Index of the root device in the signal buffer list. - _trace_name (
StaticString): Trace name used for profiling and debugging. - peer_rank (
Optional[Int]): THIS GPU's index in the signal buffer list;Nonetakes it from the device id.
Args:
- output (
T): Output tensor for THIS GPU. - input (
T): The root GPU's tensor, read through its stable slot. - signal_buffers (
TensorTuple[*Sigs.values]): Signal buffers for ALL participating GPUs. - builder (
LegacyDeviceGraphBuilder): Device graph builder for THIS GPU.