Vai al contenuto

prism_collective_core

prism_collective_core is the production-independent composition boundary. It classifies physical-port frames and connects the SINGLE control, reduction and result paths without embedding an FPGA-shell interface.

Contracts

  • Per-port ingress AXI streams: accept complete Ethernet frames from the deployment adapter.
  • Per-port egress AXI streams: publish complete CONTROL or RESULT frames and their logical byte lengths.
  • Local configuration: receive the local MAC vector and the laboratory-extension enable bit.
  • Timebase: expose millisecond_tick at the wrapper boundary; this SINGLE implementation does not yet use it for lease accounting.
  • Build parameters: select one to eight 512-bit physical ports and the link MTU (see 3.3 Parameters); the tests run one and two ports.

Contract behavior

  • Routes admitted CONTROL to the lifecycle engine, and DATA from all configured ports to the active SINGLE context.
  • Keeps logical endpoint identity separate from physical ingress port, so multiple rank bindings can share one port when configured explicitly.
  • Locks one egress frame through tlast while multiplexing control responses and results under backpressure.
  • Does not forward arbitrary Ethernet traffic or implement an INTERNAL node.
  • Leaves clock generation, reset synchronization, and shell bus adaptation to the integration wrapper.