pub fn build_program_with_self_loading_miss_handler(
workgroup_size_x: u32,
slot_count: u32,
opcodes: &[OpcodeHandler],
) -> ProgramExpand description
Build the megakernel IR with a self-loading load-miss handler.
The persistent loop is extended with an opcode::LOAD_MISS handler.
When the GPU sees this opcode it scans the IO queue for an empty slot,
writes a DMA-read request, and polls until the host/runtime marks it
complete. The arg0 field of the slot is the consumer’s opaque
resource identifier; vyre does not interpret it.