PULSAR
2.0.0
Parallel Ultra-Light Systolic Array Runtime
|
gpu_malloc_s | |
icl_deque_s | |
icl_entry_s | |
icl_hash_s | |
icl_list_s | |
MPI_Request | |
MPI_Status | |
prt_callback_finish_s | Callback data for finishing a local communication |
prt_callback_queue_s | Callback data for queueing a local communication |
prt_callback_release_s | Callback data for releasing a device packet |
prt_channel_s | VDP's data channel. Implements a data link between a pair of VDPs. Identifies the source and destination VDPs by tuples. Contains a thread-safe list of data packets |
prt_config_s | PRT configuration |
prt_device_s | VSA's accelerator device. Represents a hardware accelerator. Currently synonymous with an Nvidia GPU |
prt_mapping_s | Mapping of VDPs to hardware |
prt_packet_s | VDP's data packet A packet of data transferred through VDP's channels |
prt_proxy_s | VSA's proxy |
prt_request_s | MPI communication request for a packet. Contains a packet, some info, MPI request and MPI status |
prt_thread_s | VSA's worker thread. Represents a single CPU core or a collection of cores |
prt_transfer_s | Local transfer object |
prt_vdp_s | Virtual Data Processor (VDP). Is uniquely identified by a tuple. Fires for a predefined number of cycles. Has a fixed number of input and output channels. Has a persistent local store. Has access to read-only global store |
prt_vsa_s | Virtual Systolic Array (VSA) VSA contains global informationa about the system, a local communication proxy, an array of local worker threads, and an array of local accelerator devices |
segment |