spinn_front_end_common.interface.interface_functions package¶
Submodules¶
spinn_front_end_common.interface.interface_functions.host_no_bitfield_router_compression module¶
- class spinn_front_end_common.interface.interface_functions.host_no_bitfield_router_compression.Compression(binary_path, progress_text, result_register)[source]¶
Bases:
object
Compression algorithm implementation that uses a on-chip router compressor in order to parallelise.
- Parameters:
- compress()[source]¶
Apply the on-machine compression algorithm.
- Raises:
SpinnFrontEndException – If compression fails
- spinn_front_end_common.interface.interface_functions.host_no_bitfield_router_compression.ordered_covering_compression()[source]¶
Load routing tables and compress then using the unordered Algorithm.
To the best of our knowledge this is the same algorithm as
mundy_on_chip_router_compression()
, except this one is still buildable and can be maintained.- Raises:
SpinnFrontEndException – If compression fails
- spinn_front_end_common.interface.interface_functions.host_no_bitfield_router_compression.pair_compression()[source]¶
Load routing tables and compress then using the Pair Algorithm.
See
pacman/operations/router_compressors/pair_compressor.py
which is the exact same algorithm implemented in Python.- Raises:
SpinnFrontEndException – If compression fails
spinn_front_end_common.interface.interface_functions.machine_bit_field_router_compressor module¶
- spinn_front_end_common.interface.interface_functions.machine_bit_field_router_compressor.SIZE_OF_SDRAM_ADDRESS_IN_BYTES = 148¶
Size of SDRAM allocation for addresses
- spinn_front_end_common.interface.interface_functions.machine_bit_field_router_compressor.machine_bit_field_ordered_covering_compressor(compress_as_much_as_possible=False)[source]¶
Compression with bit field and ordered covering.
- Parameters:
compress_as_much_as_possible (bool) – whether to compress as much as possible
- Returns:
where the compressors ran
- spinn_front_end_common.interface.interface_functions.machine_bit_field_router_compressor.machine_bit_field_pair_router_compressor(compress_as_much_as_possible=False)[source]¶
Compression with bit field pairing.
- Parameters:
compress_as_much_as_possible (bool) – whether to compress as much as possible
- Returns:
where the compressors ran
Module contents¶
- class spinn_front_end_common.interface.interface_functions.FindApplicationChipsUsed[source]¶
Bases:
object
Builds a set of stats on how many chips were used for application cores.
- __call__(placements)[source]¶
Finds how many application chips there were and the cost on each chip
- Parameters:
placements (Placements) – placements
- Returns:
a tuple with 4 elements.
how many chips were used
the max application cores on any given chip
the lowest number of application cores on any given chip
the average number of application cores on any given chip
- Return type:
- spinn_front_end_common.interface.interface_functions.add_command_senders(system_placements)[source]¶
Add command senders
- spinn_front_end_common.interface.interface_functions.application_finisher()[source]¶
Handles finishing the running of an application, collecting the status of the cores that the application was running on.
- Raises:
- spinn_front_end_common.interface.interface_functions.application_runner(runtime, time_threshold, run_until_complete)[source]¶
Ensures all cores are initialised correctly, ran, and completed successfully.
- Parameters:
- Raises:
- spinn_front_end_common.interface.interface_functions.chip_io_buf_clearer()[source]¶
Clears the logging output buffer of an application running on a SpiNNaker machine.
- spinn_front_end_common.interface.interface_functions.chip_io_buf_extractor()[source]¶
Extract the logging output buffers from the machine, and separates lines based on their prefix.
- spinn_front_end_common.interface.interface_functions.chip_provenance_updater(all_core_subsets)[source]¶
- spinn_front_end_common.interface.interface_functions.chip_runtime_updater(n_sync_steps)[source]¶
Updates the runtime of an application running on a SpiNNaker machine.
- Parameters:
n_sync_steps (int or None) –
- spinn_front_end_common.interface.interface_functions.compute_energy_used()[source]¶
This algorithm does the actual work of computing energy used by a simulation (or other application) running on SpiNNaker.
- Return type:
- spinn_front_end_common.interface.interface_functions.create_notification_protocol()[source]¶
Builds the notification protocol for GUI and external device interaction.
- spinn_front_end_common.interface.interface_functions.energy_provenance_reporter(power_used)[source]¶
Converts the power usage information into provenance data.
- Parameters:
power_used (PowerUsed) – The computed basic power consumption information
- spinn_front_end_common.interface.interface_functions.execute_application_data_specs()[source]¶
Execute the data specs for all non-system targets.
- spinn_front_end_common.interface.interface_functions.execute_system_data_specs()[source]¶
Execute the data specs for all system targets.
- spinn_front_end_common.interface.interface_functions.graph_binary_gatherer()[source]¶
Extracts binaries to be executed.
- Return type:
- spinn_front_end_common.interface.interface_functions.graph_data_specification_writer(placement_order=None)[source]¶
- Parameters:
placement_order (list(Placement)) – the optional order in which placements should be examined
- Return type:
- Raises:
ConfigurationException – If the DSG asks to use more SDRAM than is available.
- spinn_front_end_common.interface.interface_functions.graph_provenance_gatherer()[source]¶
Gets provenance information from the graph.
- spinn_front_end_common.interface.interface_functions.hbp_allocator(total_run_time)[source]¶
Request a machine from the HBP remote access server that will fit a number of chips.
- Parameters:
total_run_time (int) – The total run time to request
- Returns:
machine name, machine version, BMP details (if any), reset on startup flag, auto-detect BMP, SCAMP connection details, boot port, allocation controller
- Return type:
tuple(str, int, object, bool, bool, object, object, MachineAllocationController)
- Raises:
PacmanConfigurationException – If neither n_chips or n_boards provided
- spinn_front_end_common.interface.interface_functions.host_based_bit_field_router_compressor()[source]¶
Entry point when using the PACMANAlgorithmExecutor.
- Returns:
compressed routing table entries
- Return type:
- spinn_front_end_common.interface.interface_functions.insert_chip_power_monitors_to_graphs(placements)[source]¶
Adds chip power monitors into a given graph.
param Placements placements:
- spinn_front_end_common.interface.interface_functions.insert_extra_monitor_vertices_to_graphs(placements)[source]¶
Inserts the extra monitor vertices into the graph that correspond to the extra monitor cores required.
- Returns:
vertex to Ethernet connection map, list of extra_monitor_vertices, vertex_to_chip_map
- Return type:
tuple( dict(tuple(int,int),DataSpeedUpPacketGatherMachineVertex), list(ExtraMonitorSupportMachineVertex), dict(tuple(int,int),ExtraMonitorSupportMachineVertex))
- spinn_front_end_common.interface.interface_functions.load_app_images()[source]¶
Go through the executable targets and load each binary to everywhere and then send a start request to the cores that actually use it.
- spinn_front_end_common.interface.interface_functions.load_fixed_routes()[source]¶
Load a set of fixed routes onto a SpiNNaker machine.
- Parameters:
transceiver (Transceiver) –
- spinn_front_end_common.interface.interface_functions.load_sys_images()[source]¶
Go through the executable targets and load each binary to everywhere and then send a start request to the cores that actually use it.
- spinn_front_end_common.interface.interface_functions.local_tdma_builder()[source]¶
Builds a localised TDMA.
Builds a localised TDMA which allows a number of machine vertices of the same application vertex to fire at the same time. Ensures that other application vertices are not firing at the same time. Verifies if the total time required fits into the time scale factor and machine time step. Below are text diagrams to show how this works in principle.
Figure 1: bits needed to figure out time between spikes. Cores 0-4 have 2 atoms, core 5 has 1 atom:
# 0 1 2 3 4 5 # T2-[ X X # | X X # | X X # [ X X # |------| T # X X # X <- T3 T = time_between_cores T2 = time_between_phases T3 = end of TDMA (equiv of ((n_phases + 1) * T2)) cutoff = 2. n_phases = 3 max_atoms = 2
Constants etc just to get into head:
clock cycles = 200 MHz = 200 = sv->cpu_clk
1ms = 200000 for timer 1. = clock cycles
200 per microsecond
machine time step = microseconds already.
__time_between_cores = microseconds.
Figure 2: initial offset (used to try to interleave packets from other application vertices into the TDMA without extending the overall time, and trying to stop multiple packets in flight at same time).
Figure 3: bits needed to figure out time between spikes. Cores 0-4 have 2 atoms, core 5 has 1 atom:
# 0 .5 1 .5 2 .5 3 .5 4 .5 5 .5 # T2-[ X Y X Y # | X Y X Y # | X Y X Y # [ X Y X Y # |-------| T # X Y X Y # |----| T4 # T3 -> X Y T4 is the spreader between populations. X is pop0 firing, Y is pop1 firing
- spinn_front_end_common.interface.interface_functions.locate_executable_start_type()[source]¶
Discovers where applications of particular types need to be launched.
- spinn_front_end_common.interface.interface_functions.machine_generator(bmp_details, board_version, auto_detect_bmp, scamp_connection_data, reset_machine_on_start_up)[source]¶
Makes a transceiver and a machine object.
- Parameters:
bmp_details (str) – the details of the BMP connections
board_version (int) – the version of the boards being used within the machine (1, 2, 3, 4 or 5)
auto_detect_bmp (bool) – Whether the BMP should be automatically determined
scamp_connection_data (dict((int,int), str) or None) – Job.connection dict, a String SC&MP connection data or None
reset_machine_on_start_up (bool) –
allocation_controller (MachineAllocationController) – The allocation controller; in some cases, we delegate the creation of the transceiver to it.
- Returns:
Transceiver, and description of machine it is connected to
- Return type:
- spinn_front_end_common.interface.interface_functions.placements_provenance_gatherer(n_placements, placements)[source]¶
Gets provenance information from placements.
- spinn_front_end_common.interface.interface_functions.profile_data_gatherer()[source]¶
Gets all the profiling data recorded by vertices and writes it to files.
- spinn_front_end_common.interface.interface_functions.read_routing_tables_from_machine()[source]¶
Reads compressed routing tables from a SpiNNaker machine.
- Return type:
- spinn_front_end_common.interface.interface_functions.reload_dsg_regions()[source]¶
Reloads DSG regions where needed.
- spinn_front_end_common.interface.interface_functions.routing_setup()[source]¶
Initialises the routers.
Note
This does not load any routes into them.
- Parameters:
transceiver (Transceiver) –
- spinn_front_end_common.interface.interface_functions.routing_table_loader(router_tables)[source]¶
Loads routes into initialised routers.
- Parameters:
router_tables (MulticastRoutingTables) –
- spinn_front_end_common.interface.interface_functions.spalloc_allocator(bearer_token: Optional[str] = None) Tuple[str, int, None, bool, bool, Dict[Tuple[int, int], str], None, MachineAllocationController] [source]¶
Request a machine from a SPALLOC server that will fit the given number of chips.
- spinn_front_end_common.interface.interface_functions.split_lpg_vertices(system_placements)[source]¶
Split any LPG vertices found.
- Parameters:
app_graph (ApplicationGraph) – The application graph
machine (Machine) – the SpiNNaker machine as discovered
system_placements (Placements) – existing placements to be added to
- spinn_front_end_common.interface.interface_functions.system_multicast_routing_generator()[source]¶
Generates routing table entries used by the data-in processes with the extra monitor cores.