spinn_machine package¶
Subpackages¶
Submodules¶
spinn_machine.config_setup module¶
spinn_machine.exceptions module¶
-
exception
spinn_machine.exceptions.SpinnMachineAlreadyExistsException(item, value)[source]¶ Bases:
spinn_machine.exceptions.SpinnMachineExceptionIndicates that something already exists of which there can only be one.
Parameters: -
item¶ The item of which there is already one.
-
value¶ The value of the item.
-
-
exception
spinn_machine.exceptions.SpinnMachineException[source]¶ Bases:
ExceptionA generic exception which all other exceptions in this package extend.
-
exception
spinn_machine.exceptions.SpinnMachineInvalidParameterException(parameter, value, problem)[source]¶ Bases:
spinn_machine.exceptions.SpinnMachineExceptionIndicates that there is a problem with a parameter value.
Parameters: -
parameter¶ The name of the parameter.
-
problem¶ The problem with the setting of the parameter.
-
value¶ The value of the parameter.
-
spinn_machine.full_wrap_machine module¶
-
class
spinn_machine.full_wrap_machine.FullWrapMachine(width, height, chips=None, origin=None)[source]¶ Bases:
spinn_machine.machine.MachineParameters: Raises: SpinnMachineAlreadyExistsException – If any two chips have the same x and y coordinates
-
concentric_xys(radius, start)[source]¶ A generator that produces coordinates for concentric rings of possible chips based on the links of the chips.
No check is done to see if the chip exists. This may even produce coordinates with negative numbers
Mostly copied from: https://github.com/project-rig/rig/blob/master/rig/geometry.py
Parameters: Return type:
-
get_down_xys_by_ethernet(ethernet_x, ethernet_y)[source]¶ Yields the (x,y) coordinates of the down chips on the board with this Ethernet-enabled chip.
Note
The Ethernet chip itself can not be missing if validated.
Wrap-arounds are handled as appropriate.
This method does check if the chip actually exists.
Parameters: Returns: Yields the (x, y) of the down chips on this board.
Return type:
-
get_existing_xys_by_ethernet(ethernet_x, ethernet_y)[source]¶ Yields the (x,y)s of actual chips on the board with this Ethernet-enabled chip. Including the Ethernet-enabled chip itself.
Wrap-arounds are handled as appropriate.
This method does check if the chip actually exists.
Parameters: Returns: Yields the (x,y)s of chips on this board.
Return type:
-
get_global_xy(local_x, local_y, ethernet_x, ethernet_y)[source]¶ Converts the local (X, Y) coordinates into global (x,y) coordinates, under the assumption that they are on the board with local (0,0) at global coordinates (ethernet_x, ethernet_y).
This method does take wrap-arounds into consideration.
Warning
GIGO: This method does not check if input parameters make sense, nor does it check if there is a chip at the resulting global (x,y)
Parameters: Returns: global (x,y) coordinates of the chip
Return type:
-
get_local_xy(chip)[source]¶ Converts the x and y coordinates into the local coordinates on the board as if the Ethernet-enabled chip was at position 0,0.
This method does take wrap-arounds into consideration.
This method assumes that chip is on the machine or is a copy of a chip on the machine
Parameters: chip (Chip) – A Chip in the machine Returns: Local (x, y) coordinates. Return type: tuple(int,int)
-
get_vector(source, destination)[source]¶ Get mathematical shortest vector (x, y, z) from source to destination.
This method does not check if the chips and links it assumes to take actually exist. For example long paths along a non-wrapping edge may well travel through the missing area.
This method does take wrap-arounds into consideration as applicable.
From https://github.com/project-rig/rig/blob/master/rig/geometry.py Described in http://jhnet.co.uk/articles/torus_paths
Use the same algorithm as vector_length using the best x, y pair as minimize(x, y, 0)
Parameters: Returns: The vector
-
get_vector_length(source, destination)[source]¶ Get the mathematical length of the shortest vector (x, y, z) from source to destination.
Use the same algorithm as vector to find the best x, y pair but then is optimised to directly calculate length
This method does not check if the chips and links it assumes to take actually exist. For example long paths along a non-wrapping edge may well travel through the missing area.
This method does take wrap-arounds into consideration as applicable.
From https://github.com/project-rig/rig/blob/master/rig/geometry.py Described in http://jhnet.co.uk/articles/torus_paths
On full wrap-around machines (before minimisation) the vectors can have any of the 4 combinations of positive and negative x and y The positive one is: destination - source % dimension The negative one is: positive - dimension If source is less than dimension the negative one is the wrap around If destination is greater than source the positive one wraps
One no wrap or part wrap boards the x/y that does not wrap is just destination - source
The length of vectors where both x and y have the same sign will be max(abs(x), abs(y)). As the z direction can be used in minimisation The length of vectors where x and y have opposite signs will be abs(x) and abs(y) as these are already minimum so z is not used.
Warning
GIGO: This method does not check if input parameters make sense.
Parameters: Returns: The distance in steps
Return type:
-
get_xy_cores_by_ethernet(ethernet_x, ethernet_y)[source]¶ Yields the potential (x,y) locations and the typical number of cores of all the chips on the board with this Ethernet-enabled chip.
Includes the Ethernet-enabled chip itself.
Wrap-arounds are handled as appropriate.
Note
This method does not check if the chip actually exists, nor report the actual number of cores on this chip, as is intended to be called to create the chips.
The number of cores is based on the 1,000,000 core machine where the board where built with the the 17 core chips placed in the same location on nearly every board.
Warning
GIGO! This methods assumes that ethernet_x and ethernet_y are the local 0,0 of an existing board, within the width and height of the machine.
Parameters: Returns: Yields (x, y, n_cores) where x , y are coordinates of all the potential chips on this board, and n_cores is the typical number of cores for a chip in that position.
Return type:
-
get_xys_by_ethernet(ethernet_x, ethernet_y)[source]¶ Yields the potential x,y locations of all the chips on the board with this Ethernet-enabled chip. Including the Ethernet-enabled chip itself.
Wrap-arounds are handled as appropriate.
Note
This method does not check if the chip actually exists as is intended to be called to create the chips.
Warning
GIGO! This methods assumes that ethernet_x and ethernet_y are the local 0,0 of an existing board, within the width and height of the machine.
Parameters: Returns: Yields the (x, y) coordinates of all the potential chips on this board.
Return type:
-
multiple_48_chip_boards()[source]¶ Checks that the width and height correspond to the expected size for a multi-board machine made up of more than one 48 chip board.
The assumption is that any size machine can be supported but that only ones with an expected 48 chip board size can have more than one Ethernet-enabled chip.
Returns: True if this machine can have multiple 48 chip boards Return type: bool
-
xy_over_link(x, y, link)[source]¶ Get the potential (x,y) location of the chip reached over this link.
Wrap-arounds are handled as appropriate.
Note
This method does not check if either chip source or destination actually exists as is intended to be called to create the links.
It is the callers responsibility to check the validity of this call before making it or the validity of the result.
Warning
GIGO! This methods assumes that x and y are within the width and height of the machine, and that the link goes to another chip on the machine.
On machine without full wrap-around it is possible that this method generates (x,y) values that fall outside of the legal values including negative values, x = width or y = height.
Parameters: Returns: x and y coordinates of the chip over that link if it is valid or some fictional (x,y) if not.
Return type:
-
spinn_machine.horizontal_wrap_machine module¶
-
class
spinn_machine.horizontal_wrap_machine.HorizontalWrapMachine(width, height, chips=None, origin=None)[source]¶ Bases:
spinn_machine.machine.MachineParameters: Raises: SpinnMachineAlreadyExistsException – If any two chips have the same x and y coordinates
-
concentric_xys(radius, start)[source]¶ A generator that produces coordinates for concentric rings of possible chips based on the links of the chips.
No check is done to see if the chip exists. This may even produce coordinates with negative numbers
Mostly copied from: https://github.com/project-rig/rig/blob/master/rig/geometry.py
Parameters: Return type:
-
get_down_xys_by_ethernet(ethernet_x, ethernet_y)[source]¶ Yields the (x,y) coordinates of the down chips on the board with this Ethernet-enabled chip.
Note
The Ethernet chip itself can not be missing if validated.
Wrap-arounds are handled as appropriate.
This method does check if the chip actually exists.
Parameters: Returns: Yields the (x, y) of the down chips on this board.
Return type:
-
get_existing_xys_by_ethernet(ethernet_x, ethernet_y)[source]¶ Yields the (x,y)s of actual chips on the board with this Ethernet-enabled chip. Including the Ethernet-enabled chip itself.
Wrap-arounds are handled as appropriate.
This method does check if the chip actually exists.
Parameters: Returns: Yields the (x,y)s of chips on this board.
Return type:
-
get_global_xy(local_x, local_y, ethernet_x, ethernet_y)[source]¶ Converts the local (X, Y) coordinates into global (x,y) coordinates, under the assumption that they are on the board with local (0,0) at global coordinates (ethernet_x, ethernet_y).
This method does take wrap-arounds into consideration.
Warning
GIGO: This method does not check if input parameters make sense, nor does it check if there is a chip at the resulting global (x,y)
Parameters: Returns: global (x,y) coordinates of the chip
Return type:
-
get_local_xy(chip)[source]¶ Converts the x and y coordinates into the local coordinates on the board as if the Ethernet-enabled chip was at position 0,0.
This method does take wrap-arounds into consideration.
This method assumes that chip is on the machine or is a copy of a chip on the machine
Parameters: chip (Chip) – A Chip in the machine Returns: Local (x, y) coordinates. Return type: tuple(int,int)
-
get_vector(source, destination)[source]¶ Get mathematical shortest vector (x, y, z) from source to destination.
This method does not check if the chips and links it assumes to take actually exist. For example long paths along a non-wrapping edge may well travel through the missing area.
This method does take wrap-arounds into consideration as applicable.
From https://github.com/project-rig/rig/blob/master/rig/geometry.py Described in http://jhnet.co.uk/articles/torus_paths
Use the same algorithm as vector_length using the best x, y pair as minimize(x, y, 0)
Parameters: Returns: The vector
-
get_vector_length(source, destination)[source]¶ Get the mathematical length of the shortest vector (x, y, z) from source to destination.
Use the same algorithm as vector to find the best x, y pair but then is optimised to directly calculate length
This method does not check if the chips and links it assumes to take actually exist. For example long paths along a non-wrapping edge may well travel through the missing area.
This method does take wrap-arounds into consideration as applicable.
From https://github.com/project-rig/rig/blob/master/rig/geometry.py Described in http://jhnet.co.uk/articles/torus_paths
On full wrap-around machines (before minimisation) the vectors can have any of the 4 combinations of positive and negative x and y The positive one is: destination - source % dimension The negative one is: positive - dimension If source is less than dimension the negative one is the wrap around If destination is greater than source the positive one wraps
One no wrap or part wrap boards the x/y that does not wrap is just destination - source
The length of vectors where both x and y have the same sign will be max(abs(x), abs(y)). As the z direction can be used in minimisation The length of vectors where x and y have opposite signs will be abs(x) and abs(y) as these are already minimum so z is not used.
Warning
GIGO: This method does not check if input parameters make sense.
Parameters: Returns: The distance in steps
Return type:
-
get_xy_cores_by_ethernet(ethernet_x, ethernet_y)[source]¶ Yields the potential (x,y) locations and the typical number of cores of all the chips on the board with this Ethernet-enabled chip.
Includes the Ethernet-enabled chip itself.
Wrap-arounds are handled as appropriate.
Note
This method does not check if the chip actually exists, nor report the actual number of cores on this chip, as is intended to be called to create the chips.
The number of cores is based on the 1,000,000 core machine where the board where built with the the 17 core chips placed in the same location on nearly every board.
Warning
GIGO! This methods assumes that ethernet_x and ethernet_y are the local 0,0 of an existing board, within the width and height of the machine.
Parameters: Returns: Yields (x, y, n_cores) where x , y are coordinates of all the potential chips on this board, and n_cores is the typical number of cores for a chip in that position.
Return type:
-
get_xys_by_ethernet(ethernet_x, ethernet_y)[source]¶ Yields the potential x,y locations of all the chips on the board with this Ethernet-enabled chip. Including the Ethernet-enabled chip itself.
Wrap-arounds are handled as appropriate.
Note
This method does not check if the chip actually exists as is intended to be called to create the chips.
Warning
GIGO! This methods assumes that ethernet_x and ethernet_y are the local 0,0 of an existing board, within the width and height of the machine.
Parameters: Returns: Yields the (x, y) coordinates of all the potential chips on this board.
Return type:
-
multiple_48_chip_boards()[source]¶ Checks that the width and height correspond to the expected size for a multi-board machine made up of more than one 48 chip board.
The assumption is that any size machine can be supported but that only ones with an expected 48 chip board size can have more than one Ethernet-enabled chip.
Returns: True if this machine can have multiple 48 chip boards Return type: bool
-
xy_over_link(x, y, link)[source]¶ Get the potential (x,y) location of the chip reached over this link.
Wrap-arounds are handled as appropriate.
Note
This method does not check if either chip source or destination actually exists as is intended to be called to create the links.
It is the callers responsibility to check the validity of this call before making it or the validity of the result.
Warning
GIGO! This methods assumes that x and y are within the width and height of the machine, and that the link goes to another chip on the machine.
On machine without full wrap-around it is possible that this method generates (x,y) values that fall outside of the legal values including negative values, x = width or y = height.
Parameters: Returns: x and y coordinates of the chip over that link if it is valid or some fictional (x,y) if not.
Return type:
-
spinn_machine.json_machine module¶
-
spinn_machine.json_machine.machine_from_json(j_machine)[source]¶ Generate a model of a machine from a JSON description of that machine.
Parameters: j_machine (dict in format returned by json.load or a str representing a path to the JSON file) – JSON description of the machine Returns: The machine model. Return type: Machine
spinn_machine.no_wrap_machine module¶
-
class
spinn_machine.no_wrap_machine.NoWrapMachine(width, height, chips=None, origin=None)[source]¶ Bases:
spinn_machine.machine.MachineParameters: Raises: SpinnMachineAlreadyExistsException – If any two chips have the same x and y coordinates
-
concentric_xys(radius, start)[source]¶ A generator that produces coordinates for concentric rings of possible chips based on the links of the chips.
No check is done to see if the chip exists. This may even produce coordinates with negative numbers
Mostly copied from: https://github.com/project-rig/rig/blob/master/rig/geometry.py
Parameters: Return type:
-
get_down_xys_by_ethernet(ethernet_x, ethernet_y)[source]¶ Yields the (x,y) coordinates of the down chips on the board with this Ethernet-enabled chip.
Note
The Ethernet chip itself can not be missing if validated.
Wrap-arounds are handled as appropriate.
This method does check if the chip actually exists.
Parameters: Returns: Yields the (x, y) of the down chips on this board.
Return type:
-
get_existing_xys_by_ethernet(ethernet_x, ethernet_y)[source]¶ Yields the (x,y)s of actual chips on the board with this Ethernet-enabled chip. Including the Ethernet-enabled chip itself.
Wrap-arounds are handled as appropriate.
This method does check if the chip actually exists.
Parameters: Returns: Yields the (x,y)s of chips on this board.
Return type:
-
get_global_xy(local_x, local_y, ethernet_x, ethernet_y)[source]¶ Converts the local (X, Y) coordinates into global (x,y) coordinates, under the assumption that they are on the board with local (0,0) at global coordinates (ethernet_x, ethernet_y).
This method does take wrap-arounds into consideration.
Warning
GIGO: This method does not check if input parameters make sense, nor does it check if there is a chip at the resulting global (x,y)
Parameters: Returns: global (x,y) coordinates of the chip
Return type:
-
get_local_xy(chip)[source]¶ Converts the x and y coordinates into the local coordinates on the board as if the Ethernet-enabled chip was at position 0,0.
This method does take wrap-arounds into consideration.
This method assumes that chip is on the machine or is a copy of a chip on the machine
Parameters: chip (Chip) – A Chip in the machine Returns: Local (x, y) coordinates. Return type: tuple(int,int)
-
get_vector(source, destination)[source]¶ Get mathematical shortest vector (x, y, z) from source to destination.
This method does not check if the chips and links it assumes to take actually exist. For example long paths along a non-wrapping edge may well travel through the missing area.
This method does take wrap-arounds into consideration as applicable.
From https://github.com/project-rig/rig/blob/master/rig/geometry.py Described in http://jhnet.co.uk/articles/torus_paths
Use the same algorithm as vector_length using the best x, y pair as minimize(x, y, 0)
Parameters: Returns: The vector
-
get_vector_length(source, destination)[source]¶ Get the mathematical length of the shortest vector (x, y, z) from source to destination.
Use the same algorithm as vector to find the best x, y pair but then is optimised to directly calculate length
This method does not check if the chips and links it assumes to take actually exist. For example long paths along a non-wrapping edge may well travel through the missing area.
This method does take wrap-arounds into consideration as applicable.
From https://github.com/project-rig/rig/blob/master/rig/geometry.py Described in http://jhnet.co.uk/articles/torus_paths
On full wrap-around machines (before minimisation) the vectors can have any of the 4 combinations of positive and negative x and y The positive one is: destination - source % dimension The negative one is: positive - dimension If source is less than dimension the negative one is the wrap around If destination is greater than source the positive one wraps
One no wrap or part wrap boards the x/y that does not wrap is just destination - source
The length of vectors where both x and y have the same sign will be max(abs(x), abs(y)). As the z direction can be used in minimisation The length of vectors where x and y have opposite signs will be abs(x) and abs(y) as these are already minimum so z is not used.
Warning
GIGO: This method does not check if input parameters make sense.
Parameters: Returns: The distance in steps
Return type:
-
get_xy_cores_by_ethernet(ethernet_x, ethernet_y)[source]¶ Yields the potential (x,y) locations and the typical number of cores of all the chips on the board with this Ethernet-enabled chip.
Includes the Ethernet-enabled chip itself.
Wrap-arounds are handled as appropriate.
Note
This method does not check if the chip actually exists, nor report the actual number of cores on this chip, as is intended to be called to create the chips.
The number of cores is based on the 1,000,000 core machine where the board where built with the the 17 core chips placed in the same location on nearly every board.
Warning
GIGO! This methods assumes that ethernet_x and ethernet_y are the local 0,0 of an existing board, within the width and height of the machine.
Parameters: Returns: Yields (x, y, n_cores) where x , y are coordinates of all the potential chips on this board, and n_cores is the typical number of cores for a chip in that position.
Return type:
-
get_xys_by_ethernet(ethernet_x, ethernet_y)[source]¶ Yields the potential x,y locations of all the chips on the board with this Ethernet-enabled chip. Including the Ethernet-enabled chip itself.
Wrap-arounds are handled as appropriate.
Note
This method does not check if the chip actually exists as is intended to be called to create the chips.
Warning
GIGO! This methods assumes that ethernet_x and ethernet_y are the local 0,0 of an existing board, within the width and height of the machine.
Parameters: Returns: Yields the (x, y) coordinates of all the potential chips on this board.
Return type:
-
multiple_48_chip_boards()[source]¶ Checks that the width and height correspond to the expected size for a multi-board machine made up of more than one 48 chip board.
The assumption is that any size machine can be supported but that only ones with an expected 48 chip board size can have more than one Ethernet-enabled chip.
Returns: True if this machine can have multiple 48 chip boards Return type: bool
-
xy_over_link(x, y, link)[source]¶ Get the potential (x,y) location of the chip reached over this link.
Wrap-arounds are handled as appropriate.
Note
This method does not check if either chip source or destination actually exists as is intended to be called to create the links.
It is the callers responsibility to check the validity of this call before making it or the validity of the result.
Warning
GIGO! This methods assumes that x and y are within the width and height of the machine, and that the link goes to another chip on the machine.
On machine without full wrap-around it is possible that this method generates (x,y) values that fall outside of the legal values including negative values, x = width or y = height.
Parameters: Returns: x and y coordinates of the chip over that link if it is valid or some fictional (x,y) if not.
Return type:
-
spinn_machine.vertical_wrap_machine module¶
-
class
spinn_machine.vertical_wrap_machine.VerticalWrapMachine(width, height, chips=None, origin=None)[source]¶ Bases:
spinn_machine.machine.MachineParameters: Raises: SpinnMachineAlreadyExistsException – If any two chips have the same x and y coordinates
-
concentric_xys(radius, start)[source]¶ A generator that produces coordinates for concentric rings of possible chips based on the links of the chips.
No check is done to see if the chip exists. This may even produce coordinates with negative numbers
Mostly copied from: https://github.com/project-rig/rig/blob/master/rig/geometry.py
Parameters: Return type:
-
get_down_xys_by_ethernet(ethernet_x, ethernet_y)[source]¶ Yields the (x,y) coordinates of the down chips on the board with this Ethernet-enabled chip.
Note
The Ethernet chip itself can not be missing if validated.
Wrap-arounds are handled as appropriate.
This method does check if the chip actually exists.
Parameters: Returns: Yields the (x, y) of the down chips on this board.
Return type:
-
get_existing_xys_by_ethernet(ethernet_x, ethernet_y)[source]¶ Yields the (x,y)s of actual chips on the board with this Ethernet-enabled chip. Including the Ethernet-enabled chip itself.
Wrap-arounds are handled as appropriate.
This method does check if the chip actually exists.
Parameters: Returns: Yields the (x,y)s of chips on this board.
Return type:
-
get_global_xy(local_x, local_y, ethernet_x, ethernet_y)[source]¶ Converts the local (X, Y) coordinates into global (x,y) coordinates, under the assumption that they are on the board with local (0,0) at global coordinates (ethernet_x, ethernet_y).
This method does take wrap-arounds into consideration.
Warning
GIGO: This method does not check if input parameters make sense, nor does it check if there is a chip at the resulting global (x,y)
Parameters: Returns: global (x,y) coordinates of the chip
Return type:
-
get_local_xy(chip)[source]¶ Converts the x and y coordinates into the local coordinates on the board as if the Ethernet-enabled chip was at position 0,0.
This method does take wrap-arounds into consideration.
This method assumes that chip is on the machine or is a copy of a chip on the machine
Parameters: chip (Chip) – A Chip in the machine Returns: Local (x, y) coordinates. Return type: tuple(int,int)
-
get_vector(source, destination)[source]¶ Get mathematical shortest vector (x, y, z) from source to destination.
This method does not check if the chips and links it assumes to take actually exist. For example long paths along a non-wrapping edge may well travel through the missing area.
This method does take wrap-arounds into consideration as applicable.
From https://github.com/project-rig/rig/blob/master/rig/geometry.py Described in http://jhnet.co.uk/articles/torus_paths
Use the same algorithm as vector_length using the best x, y pair as minimize(x, y, 0)
Parameters: Returns: The vector
-
get_vector_length(source, destination)[source]¶ Get the mathematical length of the shortest vector (x, y, z) from source to destination.
Use the same algorithm as vector to find the best x, y pair but then is optimised to directly calculate length
This method does not check if the chips and links it assumes to take actually exist. For example long paths along a non-wrapping edge may well travel through the missing area.
This method does take wrap-arounds into consideration as applicable.
From https://github.com/project-rig/rig/blob/master/rig/geometry.py Described in http://jhnet.co.uk/articles/torus_paths
On full wrap-around machines (before minimisation) the vectors can have any of the 4 combinations of positive and negative x and y The positive one is: destination - source % dimension The negative one is: positive - dimension If source is less than dimension the negative one is the wrap around If destination is greater than source the positive one wraps
One no wrap or part wrap boards the x/y that does not wrap is just destination - source
The length of vectors where both x and y have the same sign will be max(abs(x), abs(y)). As the z direction can be used in minimisation The length of vectors where x and y have opposite signs will be abs(x) and abs(y) as these are already minimum so z is not used.
Warning
GIGO: This method does not check if input parameters make sense.
Parameters: Returns: The distance in steps
Return type:
-
get_xy_cores_by_ethernet(ethernet_x, ethernet_y)[source]¶ Yields the potential (x,y) locations and the typical number of cores of all the chips on the board with this Ethernet-enabled chip.
Includes the Ethernet-enabled chip itself.
Wrap-arounds are handled as appropriate.
Note
This method does not check if the chip actually exists, nor report the actual number of cores on this chip, as is intended to be called to create the chips.
The number of cores is based on the 1,000,000 core machine where the board where built with the the 17 core chips placed in the same location on nearly every board.
Warning
GIGO! This methods assumes that ethernet_x and ethernet_y are the local 0,0 of an existing board, within the width and height of the machine.
Parameters: Returns: Yields (x, y, n_cores) where x , y are coordinates of all the potential chips on this board, and n_cores is the typical number of cores for a chip in that position.
Return type:
-
get_xys_by_ethernet(ethernet_x, ethernet_y)[source]¶ Yields the potential x,y locations of all the chips on the board with this Ethernet-enabled chip. Including the Ethernet-enabled chip itself.
Wrap-arounds are handled as appropriate.
Note
This method does not check if the chip actually exists as is intended to be called to create the chips.
Warning
GIGO! This methods assumes that ethernet_x and ethernet_y are the local 0,0 of an existing board, within the width and height of the machine.
Parameters: Returns: Yields the (x, y) coordinates of all the potential chips on this board.
Return type:
-
multiple_48_chip_boards()[source]¶ Checks that the width and height correspond to the expected size for a multi-board machine made up of more than one 48 chip board.
The assumption is that any size machine can be supported but that only ones with an expected 48 chip board size can have more than one Ethernet-enabled chip.
Returns: True if this machine can have multiple 48 chip boards Return type: bool
-
xy_over_link(x, y, link)[source]¶ Get the potential (x,y) location of the chip reached over this link.
Wrap-arounds are handled as appropriate.
Note
This method does not check if either chip source or destination actually exists as is intended to be called to create the links.
It is the callers responsibility to check the validity of this call before making it or the validity of the result.
Warning
GIGO! This methods assumes that x and y are within the width and height of the machine, and that the link goes to another chip on the machine.
On machine without full wrap-around it is possible that this method generates (x,y) values that fall outside of the legal values including negative values, x = width or y = height.
Parameters: Returns: x and y coordinates of the chip over that link if it is valid or some fictional (x,y) if not.
Return type:
-
Module contents¶
An abstraction of a SpiNNaker Machine. The main functionality is
provided by spinn_machine.Machine.
Functional Requirements¶
Create a machine which represents the current state of a machine, in terms of the available chips, cores on the chips, SDRAM available, routable links between chips and available routing entries.
Create a machine which represents an abstract ideal machine.
- There can only be one chip in the machine with given x, y coordinates
- There can only be one processor in each chip with a given processor ID
- There can only be one link in the router of each chip with a given ID
Add a chip to a given machine to represent an external device.
- A chip with the same x, y coordinates must not already exist in the machine
Add a link to a router of a given chip to represent a connection to an external device.
- A link with the given ID must not already exist in the chip
Create a representation of a multicast routing entry to be shared between modules that deal with routing entries.
Use Cases¶
Machineis returned as a representation of the current state of a machine.virtual_machine()is used to make an outline of a machine on which a simulation will be run, e.g., for placement of executables and/or finding routes between placed executables.Machineis extended to add a virtualChipon the machine representing an external peripheral connected to the machine directly via a link from a chip, so that routes can be directed to and from the external peripheral.MulticastRoutingEntryis returned in a list of entries, which indicate the current set of routing entries within a multicast routing table on a chip on the machine.MulticastRoutingEntryis sent in a list of routing entries to set up routing on a chip on the machine.
-
class
spinn_machine.Chip(x, y, n_processors, router, sdram, nearest_ethernet_x, nearest_ethernet_y, ip_address=None, tag_ids=None, down_cores=None, parent_link=None, v_to_p_map=None)[source]¶ Bases:
objectRepresents a SpiNNaker chip with a number of cores, an amount of SDRAM shared between the cores, and a router. The chip is iterable over the processors, yielding
(processor_id, processor)where:processor_idis the ID of a processorprocessoris theProcessorwithprocessor_id
Parameters: - x (int) – the x-coordinate of the chip’s position in the two-dimensional grid of chips
- y (int) – the y-coordinate of the chip’s position in the two-dimensional grid of chips
- n_processors (int) – the number of processors including monitor processors.
- router (Router) – a router for the chip
- sdram (SDRAM) – an SDRAM for the chip
- ip_address (str or None) – the IP address of the chip, or
Noneif no Ethernet attached - tag_ids (iterable(int) or None) – IDs to identify the chip for SDP can be empty to define no tags or None to allocate tag automatically based on if there is an ip_address
- nearest_ethernet_x (int or None) – the nearest Ethernet x coordinate
- nearest_ethernet_y (int or None) – the nearest Ethernet y coordinate
- down_cores (iterable(int) or None) – Ids of cores that are down for this Chip
- parent_link (int or None) – The link down which the parent chips is found in the tree of chips towards the root (boot) chip
- v_to_p_map (bytearray or None) – An array indexed by virtual core which gives the physical core number or 0xFF if no entry for the core
Raises: SpinnMachineAlreadyExistsException – If processors contains any two processors with the same
processor_id-
get_first_none_monitor_processor()[source]¶ Get the first processor in the list which is not a monitor core.
Return type: Processor or None
-
get_physical_core_id(virtual_p)[source]¶ Get the physical core ID from a virtual core ID.
Parameters: virtual_p (int) – The virtual core ID Return type: int or None if core not in map
-
get_physical_core_string(virtual_p)[source]¶ Get a string that can be appended to a core to show the physical core, or an empty string if not possible.
Parameters: virtual_p (int) – The virtual core ID Return type: str
-
get_processor_with_id(processor_id)[source]¶ Return the processor with the specified ID, or
Noneif the processor does not exist.Parameters: processor_id (int) – the ID of the processor to return Returns: the processor with the specified ID, or Noneif no such processorReturn type: Processor or None
-
ip_address¶ The IP address of the chip, or
Noneif there is no Ethernet connected to the chip.Return type: str or None
-
is_processor_with_id(processor_id)[source]¶ Determines if a processor with the given ID exists in the chip. Also implemented as
__contains__(processor_id)Parameters: processor_id (int) – the processor ID to check for Returns: Whether the processor with the given ID exists Return type: bool
-
parent_link¶ The link down which the parent is found in the tree of chips rooted at the machine root chip (probably 0, 0 in most cases). This will be
Noneif the chip information didn’t contain this value.Return type: int or None
-
sdram¶ The SDRAM associated with the chip.
Return type: SDRAM
-
class
spinn_machine.CoreSubset(x, y, processor_ids)[source]¶ Bases:
objectRepresents a subset of the cores on a SpiNNaker chip.
Parameters: -
add_processor(processor_id)[source]¶ Adds a processor ID to this subset
Parameters: processor_id (int) – A processor ID
-
intersect(other)[source]¶ Returns a new CoreSubset which is an intersect of this and the other.
Parameters: other (CoreSubset) – A second CoreSubset with possibly overlapping cores Returns: A new CoreSubset with any overlap Return type: CoreSubset
-
-
class
spinn_machine.CoreSubsets(core_subsets=None)[source]¶ Bases:
objectRepresents a group of CoreSubsets, with a maximum of one per SpiNNaker chip.
Parameters: core_subsets (iterable(CoreSubset)) – The cores for each desired chip -
add_core_subset(core_subset)[source]¶ Add a core subset to the set
Parameters: core_subset (CoreSubset) – The core subset to add
-
add_core_subsets(core_subsets)[source]¶ Merges a core subsets into this one.
Parameters: core_subsets (iterable(CoreSubset)) – the core subsets to add
-
core_subsets¶ The one-per-chip subsets.
Returns: Iterable of core subsets Return type: iterable(CoreSubset)
-
get_core_subset_for_chip(x, y)[source]¶ Get the core subset for a chip.
Parameters: Returns: The core subset of a chip, which will be empty if not added
Return type:
-
intersect(other)[source]¶ Returns a new CoreSubsets which is an intersect of this and the other.
Parameters: other (CoreSubsets) – A second CoreSubsets with possibly overlapping cores Returns: A new CoreSubsets with any overlap Return type: CoreSubsets
-
is_chip(x, y)[source]¶ Determine if the chip with coordinates (x, y) is in the subset
Parameters: Returns: True if the chip with coordinates (x, y) is in the subset
Return type:
-
is_core(x, y, processor_id)[source]¶ Determine if there is a chip with coordinates (x, y) in the subset, which has a core with the given ID in the subset
Parameters: Returns: Whether there is a chip with coordinates (x, y) in the subset, which has a core with the given ID in the subset
Return type:
-
values()[source]¶ Return type: iterable(CoreSubset)
-
-
class
spinn_machine.FixedRouteEntry(processor_ids, link_ids)[source]¶ Bases:
objectDescribes the sole entry in a SpiNNaker chip’s fixed route routing table. There is only one fixed route entry per chip.
Parameters:
-
class
spinn_machine.Link(source_x, source_y, source_link_id, destination_x, destination_y)[source]¶ Bases:
objectRepresents a directional link between SpiNNaker chips in the machine.
Parameters: - source_x (int) – The x-coordinate of the source chip of the link
- source_y (int) – The y-coordinate of the source chip of the link
- source_link_id (int) – The ID of the link in the source chip
- destination_x (int) – The x-coordinate of the destination chip of the link
- destination_y (int) – The y-coordinate of the destination chip of the link
-
class
spinn_machine.Machine(width, height, chips=None, origin=None)[source]¶ Bases:
objectA representation of a SpiNNaker Machine with a number of Chips. Machine is also iterable, providing
((x, y), chip)where:xis the x-coordinate of a chip,yis the y-coordinate of a chip,chipis the chip with the given(x, y)coordinates.
Use
machine_from_chips()andmachine_from_size()to determine the correct machine class.Parameters: Raises: SpinnMachineAlreadyExistsException – If any two chips have the same x and y coordinates
-
BOARD_48_CHIPS= [(0, 0), (0, 1), (0, 2), (0, 3), (1, 0), (1, 1), (1, 2), (1, 3), (1, 4), (2, 0), (2, 1), (2, 2), (2, 3), (2, 4), (2, 5), (3, 0), (3, 1), (3, 2), (3, 3), (3, 4), (3, 5), (3, 6), (4, 0), (4, 1), (4, 2), (4, 3), (4, 4), (4, 5), (4, 6), (4, 7), (5, 1), (5, 2), (5, 3), (5, 4), (5, 5), (5, 6), (5, 7), (6, 2), (6, 3), (6, 4), (6, 5), (6, 6), (6, 7), (7, 3), (7, 4), (7, 5), (7, 6), (7, 7)]¶
-
BOARD_VERSION_FOR_48_CHIPS= [4, 5]¶
-
BOARD_VERSION_FOR_4_CHIPS= [2, 3]¶
-
CHIPS_PER_BOARD= {(0, 0): 18, (0, 1): 18, (0, 2): 18, (0, 3): 18, (1, 0): 18, (1, 1): 17, (1, 2): 18, (1, 3): 17, (1, 4): 18, (2, 0): 18, (2, 1): 18, (2, 2): 18, (2, 3): 18, (2, 4): 18, (2, 5): 18, (3, 0): 18, (3, 1): 17, (3, 2): 18, (3, 3): 17, (3, 4): 18, (3, 5): 17, (3, 6): 18, (4, 0): 18, (4, 1): 18, (4, 2): 18, (4, 3): 18, (4, 4): 18, (4, 5): 18, (4, 6): 18, (4, 7): 18, (5, 1): 18, (5, 2): 17, (5, 3): 18, (5, 4): 17, (5, 5): 18, (5, 6): 17, (5, 7): 18, (6, 2): 18, (6, 3): 18, (6, 4): 18, (6, 5): 18, (6, 6): 18, (6, 7): 18, (7, 3): 18, (7, 4): 18, (7, 5): 18, (7, 6): 18, (7, 7): 18}¶
-
DEFAULT_MAX_CORES_PER_CHIP= 18¶
-
DEFAULT_SDRAM_BYTES= 123469792¶
-
LINK_ADD_TABLE= [(1, 0), (1, 1), (0, 1), (-1, 0), (-1, -1), (0, -1)]¶
-
MAX_BANDWIDTH_PER_ETHERNET_CONNECTED_CHIP= 2560¶
-
MAX_CHIPS_PER_48_BOARD= 48¶
-
MAX_CHIPS_PER_4_CHIP_BOARD= 4¶
-
MAX_CHIP_X_ID_ON_ONE_BOARD= 7¶
-
MAX_CHIP_Y_ID_ON_ONE_BOARD= 7¶
-
NON_USER_CORES= 1¶
-
ROUTER_ENTRIES= 1023¶
-
SIZE_X_OF_ONE_BOARD= 8¶
-
SIZE_Y_OF_ONE_BOARD= 8¶
-
add_chip(chip)[source]¶ Add a chip to the machine.
Parameters: chip (Chip) – The chip to add to the machine Raises: SpinnMachineAlreadyExistsException – If a chip with the same x and y coordinates already exists
-
add_chips(chips)[source]¶ Add some chips to the machine.
Parameters: chips (iterable(Chip)) – an iterable of chips Raises: SpinnMachineAlreadyExistsException – If a chip with the same x and y coordinates as one being added already exists
-
add_fpga_links()[source]¶ Add FPGA links that are on a given machine depending on the version of the board.
-
add_spinnaker_links()[source]¶ Add SpiNNaker links that are on a given machine depending on the version of the board.
-
chip_coordinates¶ An iterable of chip coordinates in the machine.
Return type: iterable(tuple(int,int))
-
concentric_xys(radius, start)[source]¶ A generator that produces coordinates for concentric rings of possible chips based on the links of the chips.
No check is done to see if the chip exists. This may even produce coordinates with negative numbers
Mostly copied from: https://github.com/project-rig/rig/blob/master/rig/geometry.py
Parameters: Return type:
-
cores_and_link_output_string()[source]¶ Get a string detailing the number of cores and links.
Return type: str
-
ethernet_connected_chips¶ The chips in the machine that have an Ethernet connection.
Return type: iterable(Chip)
-
get_chip_at(x, y)[source]¶ Get the chip at a specific (x, y) location. Also implemented as
__getitem__((x, y))Parameters: Returns: the chip at the specified location, or
Noneif no such chipReturn type:
-
get_chips_by_ethernet(ethernet_x, ethernet_y)[source]¶ Yields the actual chips on the board with this Ethernet-enabled chip. Including the Ethernet-enabled chip itself.
Wrap-arounds are handled as appropriate.
This method does check if the chip actually exists.
Parameters: Returns: Yields the chips on this board.
Return type: iterable(Chip)
-
get_cores_and_link_count()[source]¶ Get the number of cores and links from the machine.
Links are assumed to be bidirectional so the total links counted is half of the unidirectional links found.
SpiNNaker and FPGA links are not included.
Returns: tuple of (n_cores, n_links) Return type: tuple(int,int)
-
get_down_xys_by_ethernet(ethernet_x, ethernet_y)[source]¶ Yields the (x,y) coordinates of the down chips on the board with this Ethernet-enabled chip.
Note
The Ethernet chip itself can not be missing if validated.
Wrap-arounds are handled as appropriate.
This method does check if the chip actually exists.
Parameters: Returns: Yields the (x, y) of the down chips on this board.
Return type:
-
get_existing_xys_by_ethernet(ethernet_x, ethernet_y)[source]¶ Yields the (x,y)s of actual chips on the board with this Ethernet-enabled chip. Including the Ethernet-enabled chip itself.
Wrap-arounds are handled as appropriate.
This method does check if the chip actually exists.
Parameters: Returns: Yields the (x,y)s of chips on this board.
Return type:
-
get_existing_xys_on_board(chip)[source]¶ Get the chips that are on the same board as the given chip.
Parameters: chip – The chip to find other chips on the same board as Returns: An iterable of (x, y) coordinates of chips on the same board Return type: iterable(tuple(int,int))
-
get_fpga_link_with_id(fpga_id, fpga_link_id, board_address=None, chip_coords=None)[source]¶ Get an FPGA link data item that corresponds to the FPGA and FPGA link for a given board address.
Parameters: - fpga_id (int) – the ID of the FPGA that the data is going through. Refer to technical document located here for more detail: https://drive.google.com/file/d/0B9312BuJXntlVWowQlJ3RE8wWVE
- fpga_link_id (int) – the link ID of the FPGA. Refer to technical document located here for more detail: https://drive.google.com/file/d/0B9312BuJXntlVWowQlJ3RE8wWVE
- board_address (str or None) – optional board address that this FPGA link is associated with. This is ignored if chip_coords is not None. If this is None and chip_coords is None, the boot board will be assumed.
- chip_coords (tuple(int, int) or None) – optional chip coordinates that this FPGA link is associated with. If this is None and board_address is None, the boot board will be assumed.
Returns: the given FPGA link object or
Noneif no such linkReturn type:
-
get_global_xy(local_x, local_y, ethernet_x, ethernet_y)[source]¶ Converts the local (X, Y) coordinates into global (x,y) coordinates, under the assumption that they are on the board with local (0,0) at global coordinates (ethernet_x, ethernet_y).
This method does take wrap-arounds into consideration.
Warning
GIGO: This method does not check if input parameters make sense, nor does it check if there is a chip at the resulting global (x,y)
Parameters: Returns: global (x,y) coordinates of the chip
Return type:
-
get_local_xy(chip)[source]¶ Converts the x and y coordinates into the local coordinates on the board as if the Ethernet-enabled chip was at position 0,0.
This method does take wrap-arounds into consideration.
This method assumes that chip is on the machine or is a copy of a chip on the machine
Parameters: chip (Chip) – A Chip in the machine Returns: Local (x, y) coordinates. Return type: tuple(int,int)
-
get_spinnaker_link_with_id(spinnaker_link_id, board_address=None, chip_coords=None)[source]¶ Get a SpiNNaker link with a given ID.
Parameters: - spinnaker_link_id (int) – The ID of the link
- board_address (str or None) – optional board address that this SpiNNaker link is associated with. This is ignored if chip_coords is not None. If this is None and chip_coords is None, the boot board will be assumed.
- chip_coords (tuple(int, int) or None) – optional chip coordinates that this SpiNNaker link is associated with. If this is None and board_address is None, the boot board will be assumed.
Returns: The SpiNNaker link data or None if no link
Return type:
-
get_unused_xy()[source]¶ Finds an unused (x,y) coordinate on this machine.
This method will not return an (x,y) of an existing chip
This method will not return an (x,y) on any existing board even if that chip does not exist, i.e., it will not return (x,y) of a known dead chip.
It will however return the same unused_xy until a chip is added at that location.
Returns: an unused (x,y) coordinate Return type: (int, int)
-
get_vector(source, destination)[source]¶ Get mathematical shortest vector (x, y, z) from source to destination.
This method does not check if the chips and links it assumes to take actually exist. For example long paths along a non-wrapping edge may well travel through the missing area.
This method does take wrap-arounds into consideration as applicable.
From https://github.com/project-rig/rig/blob/master/rig/geometry.py Described in http://jhnet.co.uk/articles/torus_paths
Use the same algorithm as vector_length using the best x, y pair as minimize(x, y, 0)
Parameters: Returns: The vector
-
get_vector_length(source, destination)[source]¶ Get the mathematical length of the shortest vector (x, y, z) from source to destination.
Use the same algorithm as vector to find the best x, y pair but then is optimised to directly calculate length
This method does not check if the chips and links it assumes to take actually exist. For example long paths along a non-wrapping edge may well travel through the missing area.
This method does take wrap-arounds into consideration as applicable.
From https://github.com/project-rig/rig/blob/master/rig/geometry.py Described in http://jhnet.co.uk/articles/torus_paths
On full wrap-around machines (before minimisation) the vectors can have any of the 4 combinations of positive and negative x and y The positive one is: destination - source % dimension The negative one is: positive - dimension If source is less than dimension the negative one is the wrap around If destination is greater than source the positive one wraps
One no wrap or part wrap boards the x/y that does not wrap is just destination - source
The length of vectors where both x and y have the same sign will be max(abs(x), abs(y)). As the z direction can be used in minimisation The length of vectors where x and y have opposite signs will be abs(x) and abs(y) as these are already minimum so z is not used.
Warning
GIGO: This method does not check if input parameters make sense.
Parameters: Returns: The distance in steps
Return type:
-
get_xy_cores_by_ethernet(ethernet_x, ethernet_y)[source]¶ Yields the potential (x,y) locations and the typical number of cores of all the chips on the board with this Ethernet-enabled chip.
Includes the Ethernet-enabled chip itself.
Wrap-arounds are handled as appropriate.
Note
This method does not check if the chip actually exists, nor report the actual number of cores on this chip, as is intended to be called to create the chips.
The number of cores is based on the 1,000,000 core machine where the board where built with the the 17 core chips placed in the same location on nearly every board.
Warning
GIGO! This methods assumes that ethernet_x and ethernet_y are the local 0,0 of an existing board, within the width and height of the machine.
Parameters: Returns: Yields (x, y, n_cores) where x , y are coordinates of all the potential chips on this board, and n_cores is the typical number of cores for a chip in that position.
Return type:
-
get_xys_by_ethernet(ethernet_x, ethernet_y)[source]¶ Yields the potential x,y locations of all the chips on the board with this Ethernet-enabled chip. Including the Ethernet-enabled chip itself.
Wrap-arounds are handled as appropriate.
Note
This method does not check if the chip actually exists as is intended to be called to create the chips.
Warning
GIGO! This methods assumes that ethernet_x and ethernet_y are the local 0,0 of an existing board, within the width and height of the machine.
Parameters: Returns: Yields the (x, y) coordinates of all the potential chips on this board.
Return type:
-
is_chip_at(x, y)[source]¶ Determine if a chip exists at the given coordinates. Also implemented as
__contains__((x, y))Parameters: Returns: True if the chip exists, False otherwise
Return type:
-
local_xys¶ Provides a list of local (x,y) coordinates for a perfect board on this machine.
Local (x,y)s never include wrap-arounds.
Note
No check is done to see if any board in the machine actually has a chip with this local (x, y).
Return type: iterable(tuple(int,int))
-
static
max_cores_per_chip()[source]¶ Gets the max core per chip for the while system.
There is no guarantee that there will be any Chips with this many cores, only that there will be no cores with more.
Returns: the default cores per chip unless overridden by set
-
multiple_48_chip_boards()[source]¶ Checks that the width and height correspond to the expected size for a multi-board machine made up of more than one 48 chip board.
The assumption is that any size machine can be supported but that only ones with an expected 48 chip board size can have more than one Ethernet-enabled chip.
Returns: True if this machine can have multiple 48 chip boards Return type: bool
-
static
set_max_cores_per_chip(new_max)[source]¶ Allows setting the max number of cores per chip for the whole system.
Allows virtual machines to go higher than normal.
Real machines can only be capped never increased beyond what they actually have.
Parameters: new_max (int) – New value to use for the max Raises: SpinnMachineException – if max_cores_per_chip has already been used and is now being changed. This exception also happens if the value is set twice to different values. For example in the script and in the configuration file.
-
spinnaker_links¶ The set of SpiNNaker links in the machine.
Return type: iterable(tuple(tuple(str,int), SpinnakerLinkData))
-
total_available_user_cores¶ The total number of cores on the machine which are not monitor cores.
Return type: int
-
unreachable_incoming_chips()[source]¶ Detects chips that are not reachable from any of their neighbours.
Current implementation does not deal with group of unreachable chips.
Returns: List (hopefully empty) if the (x,y) coordinates of unreachable chips. Return type: list(tuple(int,int))
-
unreachable_incoming_local_chips()[source]¶ Detects chips that are not reachable from any of their local neighbours.
Current implementation does not deal with group of unreachable chips.
Returns: List (hopefully empty) if the (x,y) coordinates of unreachable chips. Return type: list(tuple(int,int))
-
unreachable_outgoing_chips()[source]¶ Detects chips that can not reach any of their neighbours.
Current implementation does not deal with group of unreachable chips.
Returns: List (hopefully empty) if the (x,y) coordinates of unreachable chips. Return type: list(tuple(int,int))
-
unreachable_outgoing_local_chips()[source]¶ Detects chips that can not reach any of their local neighbours.
Current implementation does not deal with group of unreachable chips.
Returns: List (hopefully empty) if the (x,y) coordinates of unreachable chips. Return type: list(tuple(int,int))
-
validate()[source]¶ Validates the machine and raises an exception in unexpected conditions.
Assumes that at the time this is called all chips are on the board.
This allows the checks to be avoided when creating a virtual machine (Except of course in testing)
Raises: SpinnMachineException – - An Error is raised if there is a chip with a x outside of the range 0 to width -1.
- An Error is raised if there is a chip with a y outside of the range 0 to height -1.
- An Error is raise if there is no chip at the declared Ethernet-enabled chip x and y.
- An Error is raised if an Ethernet-enabled chip is not at a local 0,0.
- An Error is raised if there is no Ethernet-enabled chip is at 0,0.
- An Error is raised if this is a unexpected multiple board situation.
-
where_is_chip(chip)[source]¶ Returns global and local location for this chip.
This method assumes that chip is on the machine or is a copy of a chip on the machine.
Parameters: chip (Chip) – A Chip in the machine Returns: A human-readable description of the location of a chip. Return type: str
-
where_is_xy(x, y)[source]¶ Returns global and local location for this chip.
Parameters: Returns: A human-readable description of the location of a chip.
Return type:
-
xy_over_link(x, y, link)[source]¶ Get the potential (x,y) location of the chip reached over this link.
Wrap-arounds are handled as appropriate.
Note
This method does not check if either chip source or destination actually exists as is intended to be called to create the links.
It is the callers responsibility to check the validity of this call before making it or the validity of the result.
Warning
GIGO! This methods assumes that x and y are within the width and height of the machine, and that the link goes to another chip on the machine.
On machine without full wrap-around it is possible that this method generates (x,y) values that fall outside of the legal values including negative values, x = width or y = height.
Parameters: Returns: x and y coordinates of the chip over that link if it is valid or some fictional (x,y) if not.
Return type:
-
class
spinn_machine.MulticastRoutingEntry(routing_entry_key, mask, processor_ids=None, link_ids=None, defaultable=False, spinnaker_route=None)[source]¶ Bases:
objectRepresents an entry in a SpiNNaker chip’s multicast routing table. There can be up to 1024 such entries per chip, but some may be reserved for system purposes.
Note
The processor_ids and link_ids parameters are only optional if a spinnaker_route is provided. If a spinnaker_route is provided the processor_ids and link_ids parameters are ignored.
Parameters: - routing_entry_key (int) – The routing key_combo
- mask (int) – The route key_combo mask
- processor_ids (iterable(int) or None) – The destination processor IDs
- link_ids (iterable(int) or None) – The destination link IDs
- defaultable (bool) – If this entry is defaultable (it receives packets from its directly opposite route position)
- spinnaker_route (int or None) – The processor_ids and link_ids expressed as a single int.
Raises: - spinn_machine.exceptions.SpinnMachineAlreadyExistsException –
- If processor_ids contains the same ID more than once
- If link_ids contains the same ID more than once
- TypeError – if no spinnaker_route provided and either processor_ids or link_ids is missing or None
-
defaultable¶ Whether this entry is a defaultable entry. An entry is defaultable if it is duplicating the default behaviour of the SpiNNaker router (to pass a message out on the link opposite from where it was received, without routing it to any processors; source and destination chips for a message cannot be defaultable).
Return type: bool
-
merge(other_entry)[source]¶ Merges together two multicast routing entries. The entry to merge must have the same key and mask. The merge will join the processor IDs and link IDs from both the entries. This could be used to add a new destination to an existing route in a routing table. It is also possible to use the add (+) operator or the or (|) operator with the same effect.
Parameters: other_entry (MulticastRoutingEntry) – The multicast entry to merge with this entry Returns: A new multicast routing entry with merged destinations Return type: MulticastRoutingEntry Raises: spinn_machine.exceptions.SpinnMachineInvalidParameterException – If the key and mask of the other entry do not match
-
class
spinn_machine.Processor(processor_id, clock_speed=200000000, is_monitor=False, dtcm_available=65536)[source]¶ Bases:
objectA processor object included in a SpiNNaker chip.
Parameters: - processor_id (int) – ID of the processor in the chip
- clock_speed (int) – The number of CPU cycles per second of the processor
- is_monitor (bool) – Determines if the processor is considered the monitor processor, and so should not be otherwise allocated
- dtcm_available (int) – Data Tightly Coupled Memory available
Raises: spinn_machine.exceptions.SpinnMachineInvalidParameterException – If the clock speed is negative
-
CLOCK_SPEED= 200000000¶
-
DTCM_AVAILABLE= 65536¶
-
cpu_cycles_available¶ The number of CPU cycles available from this processor per ms.
Return type: int
-
class
spinn_machine.Router(links, emergency_routing_enabled=False, n_available_multicast_entries=1023)[source]¶ Bases:
objectRepresents a router of a chip, with a set of available links. The router is iterable over the links, providing (source_link_id, link) where:
source_link_idis the ID of a linklinkis theLinkwith IDsource_link_id
Parameters: Raises: SpinnMachineAlreadyExistsException – If any two links have the same
source_link_id-
LINK_OPPOSITE= 3¶
-
MAX_CORES_PER_ROUTER= 18¶
-
MAX_LINKS_PER_ROUTER= 6¶
-
add_link(link)[source]¶ Add a link to the router of the chip.
Parameters: link (Link) – The link to be added Raises: SpinnMachineAlreadyExistsException – If another link already exists with the same source_link_id
-
static
convert_routing_table_entry_to_spinnaker_route(routing_table_entry)[source]¶ Convert a routing table entry represented in software to a binary routing table entry usable on the machine.
Parameters: routing_table_entry (MulticastRoutingEntry or FixedRouteEntry) – The entry to convert Return type: int
-
static
convert_spinnaker_route_to_routing_ids(route)[source]¶ Convert a binary routing table entry usable on the machine to lists of route IDs usable in a routing table entry represented in software.
Parameters: route (int) – The routing table entry Returns: The list of processor IDs, and the list of link IDs. Return type: tuple(list(int), list(int))
-
get_link(source_link_id)[source]¶ Get the link with the given ID, or None if no such link. Also implemented as
__getitem__(source_link_id)Parameters: source_link_id (int) – The ID of the link to find Returns: The link, or Noneif no such linkReturn type: Link or None
-
get_neighbouring_chips_coords()[source]¶ Utility method to convert links into x and y coordinates.
Returns: iterable list of destination coordinates in x and y dict Return type: iterable(dict(str,int))
-
is_link(source_link_id)[source]¶ Determine if there is a link with ID source_link_id. Also implemented as
__contains__(source_link_id)Parameters: source_link_id (int) – The ID of the link to find Returns: True if there is a link with the given ID, False otherwise Return type: bool
-
class
spinn_machine.SpiNNakerTriadGeometry(triad_width, triad_height, roots, centre)[source]¶ Bases:
objectGeometry of a “triad” of SpiNNaker boards.
The geometry is defined by the arguments to the constructor; the standard arrangement can be obtained from get_spinn5_geometry.
Note
The geometry defines what a Triad is in terms of the dimensions of a triad and where the Ethernet chips occur in the triad.
Parameters: -
get_ethernet_chip_coordinates(x, y, width, height, root_x=0, root_y=0)[source]¶ Get the coordinates of a chip’s local Ethernet connected chip according to this triad geometry object.
Warning
local_eth_coord()will always produce the coordinates of the Ethernet-connected SpiNNaker chip on the same SpiNN-5 board as the supplied chip. This chip may not actually be working.Parameters: - x (int) – x-coordinate of the chip to find the nearest Ethernet of
- y (int) – y-coordinate of the chip to find the nearest Ethernet of
- width (int) – width of the SpiNNaker machine (must be a multiple of the triad width of this geometry)
- height (int) – height of the SpiNNaker machine (must be a multiple of the triad height of this geometry)
- root_x (int) – x-coordinate of the boot chip (default 0, 0)
- root_y (int) – y-coordinate of the boot chip (default 0, 0)
Returns: The coordinates of the closest Ethernet chip
Return type:
-
get_local_chip_coordinate(x, y, root_x=0, root_y=0)[source]¶ Get the coordinates of a chip on its board of a multi-board system relative to the Ethernet chip of the board.
Note
This function assumes the system is constructed from SpiNN-5 boards
Parameters: Returns: the coordinates of the chip relative to its board
Return type:
-
get_potential_ethernet_chips(width, height)[source]¶ Get the coordinates of chips that should be Ethernet chips
Parameters: Return type:
-
static
get_spinn5_geometry()[source]¶ Get the geometry object for a SpiNN-5 arrangement of boards
Returns: a SpiNNakerTriadGeometryobject.
-
spinn5_triad_geometry= None¶
-
-
spinn_machine.virtual_machine(width, height, n_cpus_per_chip=None, validate=True)[source]¶ Create a virtual SpiNNaker machine, used for planning execution.
Parameters: Returns: a virtual machine (that cannot execute code)
Return type:
-
spinn_machine.machine_from_chips(chips)[source]¶ Create a machine with the assumed wrap-around based on the sizes.
The size of the machine is calculated from the list of chips.
Parameters: chips (list(Chip)) – Full list of all chips on this machine. Or at least a list which includes a chip with the highest X and one with the highest Y (excluding any virtual chips) Returns: A subclass of Machine Return type: Machine
-
spinn_machine.machine_from_size(width, height, chips=None, origin=None)[source]¶ Create a machine with the assumed wrap-around based on the sizes.
This could include a machine with no wrap-arounds, only vertical ones, only horizontal ones or both.
Note
If the sizes do not match the ones for a known wrap-around machine, a machine with no wrap-arounds is assumed.
Parameters: - width (int) – The width of the machine excluding any virtual chips
- height (int) – The height of the machine excluding any virtual chips
- chips (list(Chip) or None) – Any chips to be added.
- origin (str or None) – Extra information about how this machine was created
to be used in the str method. Example “
Virtual” or “Json”
Returns: A subclass of Machine
Return type: