Split cores by rowleya · Pull Request #1558 · SpiNNakerManchester/sPyNNaker

rowleya · 2025-05-02T15:33:46Z

This integrates the execution of neuron code and synapse code on separate cores more completely, making it nicer for the user to use. This includes:

The user can choose to set a number of synapse cores for a model (like the number of neurons per core)
The user decide whether they want to allow delay extensions or not for a model (as above)
The user can control the number of neurons per core, synapse cores and allowing of delay extensions when creating populations.
The user can use 0.1ms timesteps in realtime without having to add an option to the configs.
The code will try to work out whether to use synapse cores and how many synapse cores to use if so, and whether to allow delay extensions.

What it doesn't do:

Theoretically, the use of synapse cores shouldn't need to depend on the timestep. In this code, it won't automatically use synapse cores unless the timestep is less than 1.0. See Allow detection of merged synapse core vs. split cores at any time step #1559.

Requires:

Tested by:

Test split_cores IntegrationTests#317

Christian-B

remove one _n_synapse_cores and one _allow_delay_extensions

spynnaker/pyNN/data/spynnaker_data_view.py

spynnaker/pyNN/models/neuron/abstract_pynn_neuron_model.py

…sPyNNaker.git into split_cores

Christian-B · 2025-05-08T06:04:49Z

spynnaker/pyNN/spinnaker.py

            db.insert_version("lazyarray_version", lazyarray_version)

+        # Clears all previously added ceiling on the number of neurons per core
+        AbstractPyNNModel.reset_all()


Not needed as also called by AbstractPyNNNeuronModel.reset_all()

Yes, I wasn't certain if that actually did the right thing; I guess that is true though!

spynnaker/pyNN/models/abstract_models/sends_synaptic_inputs_over_sdram.py

spynnaker/pyNN/extra_algorithms/splitter_components/spynnaker_splitter_selector.py

spynnaker/pyNN/models/neuron/population_vertex.py

Christian-B · 2025-05-08T10:22:19Z

spynnaker/pyNN/models/neuron/synapse_dynamics/abstract_synapse_dynamics.py

+    @abstractmethod
+    def synapses_per_second(self) -> int:
+        """
+        Approximate number of synapses that can be processed per second


Would be good to add a comment about which it does under estimate, over estimate or as close as possible.

Should that not be the other way around. Underestimating will create a system that can not handle the received spikes

No, if you underestimate how many spikes you can process, you can process more than you estimate - that is, your estimate is under what you can really handle. If you overestimate you are guessing you can do more than you really can, so more spikes will be sent to you than you can handle.

spynnaker/pyNN/models/neuron/synapse_dynamics/abstract_synapse_dynamics.py

spynnaker/pyNN/models/spike_source/spike_source_poisson_machine_vertex.py

Christian-B

can be merged when green

rowleya added 30 commits April 4, 2025 11:57

Try to automate the splitting

ae3d643

Flake8

cf28997

Merge branch 'master' into split_cores

463632d

Pylint fixes

3a5053d

Don't forget the Poisson cores (and remove the commented out ones)

8c6d649

More flake8

a1530c5

Type fixes

a496714

More types

b8773c1

Fix so that Poisson happens first

1f5c56c

Need a return type

2e41883

This is working in theory but needs some more checking

4d04ad2

Allow a time scale factor of 1 for 0.1ms timesteps by default

e115143

Add the right function here

7aeb8c1

Flake8

5cc6efc

Allow changes to per-core things, and update elsewhere too

3241394

Fix slots

06cbd9b

pylint

8f09039

Add missing arguments

86a5b9b

Fix and remove tests

4465a9a

Make still abstract

7ab8e93

Try a single abstract method...

bb8a3be

flake8

0838b76

Types

b18299b

flake8

5da4164

More type fixes

dfba54f

flake8

98fdc84

This doesn't override...

22ea195

We allow delay extensions if needed!

5f08f20

Fix tests

bea4cda

Fix tests

7d8b48e

rowleya added 5 commits May 2, 2025 08:48

Reset parameters when something gets added

4bbeebb

Ensure the delay slots are a power of 2!

6a6b0fe

Add synapses_per_second as required

d9a401b

Speeling

c7230ca

Merge branch 'master' into split_cores

b3137f9

rowleya requested a review from Christian-B May 6, 2025 07:04

rowleya marked this pull request as ready for review May 6, 2025 07:04

rowleya mentioned this pull request May 6, 2025

Allow detection of merged synapse core vs. split cores at any time step #1559

Open

This was referenced May 6, 2025

Split cores SpiNNakerManchester/PACMAN#599

Merged

Split cores SpiNNakerManchester/SpiNNFrontEndCommon#1284

Merged

Christian-B requested changes May 6, 2025

View reviewed changes

spynnaker/pyNN/data/spynnaker_data_view.py Outdated Show resolved Hide resolved

spynnaker/pyNN/models/neuron/abstract_pynn_neuron_model.py Show resolved Hide resolved

rowleya added 6 commits May 6, 2025 16:31

Fixes to avoid multiple data locations

d121e45

Merge branch 'split_cores' of https://github.com/SpiNNakerManchester/…

5fe438c

…sPyNNaker.git into split_cores

Flake8 and update methods to actual ones

e033052

Disallow changes after pops created, and move reset to setup

4ff022b

Add docs

1967d4d

Get the type right

ba6edf5

Christian-B mentioned this pull request May 8, 2025

logger warn of sim,.setup() #1488

Closed

Christian-B requested changes May 8, 2025

View reviewed changes

rowleya added 2 commits May 8, 2025 13:48

Review changes

6e35547

Flake8

fa9ba4a

Christian-B mentioned this pull request May 8, 2025

no need to cache which types have set neurons per core #1560

Closed

Christian-B approved these changes May 8, 2025

View reviewed changes

This was referenced May 8, 2025

Split cores SpiNNakerManchester/microcircuit_model#44

Merged

Split cores SpiNNakerManchester/PyNNExamples#127

Merged

rowleya added 3 commits May 8, 2025 15:54

Spelling

7d4cf32

Type fix

150b9b4

Another type fix

05cb1d4

rowleya merged commit ad93cf0 into master May 9, 2025
13 checks passed

rowleya deleted the split_cores branch May 9, 2025 06:34

Conversation

rowleya commented May 2, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Christian-B left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Christian-B May 8, 2025

Choose a reason for hiding this comment

Uh oh!

rowleya May 8, 2025

Choose a reason for hiding this comment

Uh oh!

rowleya May 8, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Christian-B May 8, 2025

Choose a reason for hiding this comment

Uh oh!

rowleya May 8, 2025

Choose a reason for hiding this comment

Uh oh!

Christian-B May 8, 2025

Choose a reason for hiding this comment

Uh oh!

rowleya May 8, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Christian-B left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

rowleya commented May 2, 2025 •

edited

Loading