Struct anira_engine_edge#
-
struct anira_engine_edge#
How a custom engine passes one slot to its runtime, as the plan report states it (anira_plan_slot.edge_class and reason): what anira_engine_edge_fn writes. Tier 2, struct_size first, no implicit padding (LP64 16 bytes, ILP32 12). The mover follows from the class: a registered engine is handed the chunk’s buffer itself, so whatever moves the bytes is the engine’s code, ANIRA_MOVER_ENGINE for any class but ANIRA_EDGE_ZERO_COPY and ANIRA_MOVER_NONE for that.
Public Members
-
uint32_t struct_size#
sizeof(anira_engine_edge) of the library, as anira hands the record over; the engine writes within it.
-
uint32_t edge_class#
anira_edge_class: where the engine takes the slot’s bytes, from the slot’s memory until it computes on them and back for an output. Arrives as the provider rule’s for the slot’s direction (the same class both ways): ANIRA_EDGE_ZERO_COPY on ANIRA_PROVIDER_CPU and XNNPACK; on any other provider ANIRA_EDGE_UNIFIED_TRANSFER where its device computes in unified memory, ANIRA_EDGE_DEDICATED_TRANSFER where it has its own memory and ANIRA_EDGE_UNKNOWN where no probe established which (the plan’s anira_plan_info.memory_kind). An engine that leaves it states that.
-
const char *reason#
Why the edge is what it is; NULL with nothing to say (arrives as the provider rule’s for the slot’s direction: a device provider’s names the upload of an input or the readback of an output). Copied before the callback returns.
-
uint32_t struct_size#