3. PopVision Analysis library Python API
This document describes the libpva Python API. Many classes are wrappers around the equivalent C++ class.
3.1. Compilation reports
- pva.openReport(*args, **kwargs)
Overloaded function.
openReport(filename: str) -> pva::Report
openReport(profileFilename: str, debugInfoFilename: str) -> pva::Report
- class pva.Report
- property compilation
Details from the Poplar compilation.
- property execution
Details from the Poplar execution.
- Type
- property instrumentationSettings
Details of the instrumentation settings used.
- listPrintLength(self: pva.pva_core.Report, newLength: int) None
- property poplarVersion
The Poplar version.
- Type
- property version
The report format version.
- class pva.PoplarVersion
The Poplar version that was used when generating the report.
- class pva.ProfileReportVersion
ProfileReportVersion contains the version of the report format that has been used.
- class pva.InstrumentationSettings
Information about the instrumentation settings used when compiling the graph.
- class ComputeInstrumentationLevel
Members:
Off
Vertex
Tile
Ipu
Device
Unknown
- Device = <ComputeInstrumentationLevel.Device: 4>
- Ipu = <ComputeInstrumentationLevel.Ipu: 3>
- Off = <ComputeInstrumentationLevel.Off: 0>
- Tile = <ComputeInstrumentationLevel.Tile: 2>
- Unknown = <ComputeInstrumentationLevel.Unknown: 5>
- Vertex = <ComputeInstrumentationLevel.Vertex: 1>
- property name
- property value
- Device = <ComputeInstrumentationLevel.Device: 4>
- class ExternalExchangeInstrumentationLevel
Members:
Off
Tile
Unknown
- Off = <ExternalExchangeInstrumentationLevel.Off: 0>
- Tile = <ExternalExchangeInstrumentationLevel.Tile: 1>
- Unknown = <ExternalExchangeInstrumentationLevel.Unknown: 2>
- property name
- property value
- Ipu = <ComputeInstrumentationLevel.Ipu: 3>
- Off = <ExternalExchangeInstrumentationLevel.Off: 0>
- Tile = <ExternalExchangeInstrumentationLevel.Tile: 1>
- Unknown = <ExternalExchangeInstrumentationLevel.Unknown: 2>
- Vertex = <ComputeInstrumentationLevel.Vertex: 1>
- property compute
Compute instrumentation level.
- Type
pva.ComputeInstrumentationLevel
- property externalExchange
External exchange instrumentation level.
- Type
pva.ExternalExchangeInstrumentationLevel
- class pva.Graph
Basic details about the graph.
- class pva.Target
Target contains information about the target hardware.
- Cpu = <Type.Cpu: 2>
- Ipu = <Type.Ipu: 0>
- IpuModel = <Type.IpuModel: 1>
- class Type
The type of hardware
Members:
Ipu
IpuModel
Cpu
Unknown
- Cpu = <Type.Cpu: 2>
- Ipu = <Type.Ipu: 0>
- IpuModel = <Type.IpuModel: 1>
- Unknown = <Type.Unknown: 3>
- property name
- property value
- Unknown = <Type.Unknown: 3>
- property architecture
The IPU architecture.
- Type
- property atomicStoreGranularity
The atomic store granularity. Will return 0 if this data is not present in the report
- Type
- property convUnitCoeffLoadBytesPerCycle
The number of convolutional weights that can be loaded in a cycle. Will return 0 if this data is not present in the report
- Type
- property dataPathWidth
The width of the load/store data path within the tile. Will return 0 if this data is not present in the report
- Type
- property exchangeBytesPerCycle
The bandwidth of internal IPU exchange in bytes per cycle. Will return 0 if this data is not present in the report
- Type
- property fp16ConvUnitInputLoadElemsPerCycle
The input elements loaded per cycle for f16 conv. Will return 0 if this data is not present in the report
- Type
- property fp16ConvUnitMaxPipelineDepth
The maximum pipeline depth of the convolution units within tile for fp16. Will return 0 if this data is not present in the report
- Type
- property fp16InFp16OutConvUnitsPerTile
The number of convolution units in the tile that can be used when partial results are outputs as 16-bits and inputs are 16 bits. Will return 0 if this data is not present in the report
- Type
- property fp16InFp32OutConvUnitsPerTile
The number of convolution units in the tile that can be used when partial results are outputs as 32-bits and inputs are 16 bits. Will return 0 if this data is not present in the report
- Type
- property fp32ConvUnitInputLoadElemsPerCycle
The input elements loaded per cycle for f32 conv. Will return 0 if this data is not present in the report
- Type
- property fp32ConvUnitMaxPipelineDepth
The maximum pipeline depth of the convolution units within tile for fp32 Will return 0 if this data is not present in the report.
- Type
- property fp32InFp32OutConvUnitsPerTile
The number of convolution units in the tile that can be used when partial results are outputs as 32-bits and inputs are 32 bits. Will return 0 if this data is not present in the report
- Type
- property fp8ConvUnitInputLoadElemsPerCycle
The input elements loaded per cycle for f8 conv. Will return 0 if this data is not present in the report
- Type
- property fp8ConvUnitMaxPipelineDepth
The maximum pipeline depth of the convolution units within tile for fp8. Will return 0 if this data is not present in the report
- Type
- property fp8InFp16OutConvUnitsPerTile
The number of convolution units in the tile that can be used when partial results are outputs as 16-bits and inputs are 8 bits. Will return 0 if this data is not present in the report
- Type
- property globalExchangePacketBytes
Size of the packet used to transfer data between tiles in bytes. Will return 0 if this data is not present in the report
- Type
- property globalSyncCycles
The number of clock cycles required to synchronize all IPUs. Will return 0 if this data is not present in the report
- Type
- property instructionBytes
The size of an instruction in bytes. Will return 0 if this data is not present in the report
- Type
- property maxImmediateOffsetInRunInstr
Max range of immediate operand in run instruction zimm16 operand multiplied implicitly by 4 when added to register operand. Will return 0 if this data is not present in the report
- Type
- property memcpyBytesPerCycle
The number of bytes per cycle that can be copied from one location to another using a memcpy. Will return 0 if this data is not present in the report
- Type
- property memoryElementOffsets
std::vector<int> Offsets for each memory element.
- property memoryRegionOffsets
std::vector<int> Offsets for each memory region.
- property numStrideBits
Number of stride bits. Will return 0 if this data is not present in the report
- Type
- property numTiles
The total number of tiles. This is the product of numIPUs and tilesPerIPU. It is stored redundantly for convenience.
- Type
- property numWorkerContexts
The number of worker contexts per tile. Will return 0 if this data is not present in the report
- Type
- property rptCountMax
Max number of repeat counter. Will return 0 if this data is not present in the report
- Type
- property supportsSuperTileSendReceive
Whether a tile in a super-tile can use all the exchange bandwidth of the super-tile to send or receive, when the other tile is idle or receiving the same data. Will return false if this data is not present in the report
- Type
- property tileLocalSyncExitDelay
Number of cycles after a worker has issued its exit instruction that the supervisor can resume. Will return 0 if this data is not present in the report
- Type
- property tileLocalSyncSyncDelay
Number of cycles from issuing a sync instruction to the earliest time that instructions can resume. Will return 0 if this data is not present in the report
- Type
- property tilesPerSuperTile
The number of tiles per super-tile. Will return 0 if this data is not present in the report
- Type
- property totalMemory
The total memory. This is the product of bytesPerTile and numTiles (or bytesPerIPU and numIPUs).
- Type
- property type
The target typ.
- Type
- class pva.VertexType
VertexType represents a vertex type.
- Asm = <Source.Asm: 0>
- CPlusPlus = <Source.CPlusPlus: 1>
- class Source
Indicates how the vertex has been implemented.
Members:
Asm : Implemented in assembler.
CPlusPlus : Implemented in C++.
Unknown : Implementation unknown.
- Asm = <Source.Asm: 0>
- CPlusPlus = <Source.CPlusPlus: 1>
- Unknown = <Source.Unknown: 2>
- property name
- property value
- Unknown = <Source.Unknown: 2>
- property size
The size of the vertex type in bytes.
Contains the size of the vertex state (the class members) of each vertex type. For example Doubler might have 4 bytes of state.
- Type
- property source
How the vertex type has been implemented.
- class pva.VertexMemory
VertexMemory represents how much memory a vertex type uses on a tile.
- property type
The type of vertex.
- Type
- class pva.VertexInstances
VertexInstances represents a number of instances of a vertex type in a compute set.
- property type
The type of the vertex instances.
- Type
- class pva.MemoryOverlap
MemoryOverlap represents how much memory within a category / region is overlapped or not overlapped.
The memory used by some variables can be overlapped with others, because they are not live at the same time. Hence, the usage is split into overlapped and non-overlapped components.
- class pva.CategoryMemory
CategoryMemory represents the breakdown of memory by region for a category.
Category is a breakdown of memory usage across the whole system by the type of data, and the region it is in.
There are two memory regions on each tile, interleaved and non-interleaved, the use of each of these is reported separately. If the memory requirement is greater than the available memory, then this is reported as overflowed.
- property interleaved
The interleaved memory region.
- Type
- property nonInterleaved
The non interleaved memory region
- Type
- property overflowed
The overflowed memory region.
- Type
- property total
The sum of interleaved, non interleaved, and overflowed memory.
- Type
- class pva.TileMemoryCategories
Memory represent the memory layout of a tile in various different ways.
The memory object contains a lot of information about memory use. All memory is statically allocated so you don’t need to run the program to gather this data.
- property constant
Constants memory category.
Constants added by the user. Variables added by the compiler that happen to be constant will be categorised as “variable”.
- Type
- property controlCode
Control code memory category.
Code for Program objects and running compute sets.
- Type
- property controlId
Control id memory category.
Variables that are used in switch programs or variables that store a sync ID for tracking host/device synchronisation points.
- Type
- property controlTable
Control table memory category.
A table that lists the vertices to run in each compute set. Only used if the table scheduler is enabled.
- Type
- property copyDescriptor
Copy descriptor memory category.
Copy descriptors are special variable-sized fields used by copy vertices.
- Type
- property dwarf
DWARF memory category.
DWARF debug information.
- Type
- property globalExchangeCode
Global exchange code memory category.
Code for performing exchange operations between IPUs.
- Type
- property globalExchangePacketHeader
Global exchange packet header category.
Packet headers for exchange operations between IPUs.
- Type
- property globalMessage
Global message memory category.
Message variables holding data being sent between IPUs.
- Type
- property hostExchangeCode
Host exchange code memory category.
Code for performing exchange operations to and from the host
- Type
- property hostExchangePacketHeader
Host exchange packet header memory category.
Data used as packet headers for host exchange.
- Type
- property hostMessage
Host message memory category.
Message variables holding data being sent or received from the host.
- Type
- property instrumentationResults
Instrumentation results memory category.
Storage for profiling information.
- Type
- property internalExchangeCode
Internal exchange code memory category.
Code for performing internal exchanges.
- Type
- property message
Message memory category.
Message data for internal exchanges.
- Type
- property multiple
Multiple memory category.
Space shared by variables from multiple different categories.
- Type
- property outputEdge
Output edge memory category.
Storage for output edge data before an exchange takes place.
- Type
- property rearrangement
Rearrangement memory category.
Variables holding rearranged versions of tensor data. A rearranged variable will never be always live as it is only required in the context of a specific compute set.
- Type
Shared code storage memory category.
Code shared by vertices.
- Type
Shared data storage memory category.
Data shared by vertices.
- Type
- property stack
Stack memory category.
The work and supervisor stacks allocated on the specified tile. For more information about worker stack allocation see the Vertex Assembly Programming Guide.
- Type
- property variable
Variable memory category.
Variables created in the program (for example, created by the Poplar graph.addVariable() function).
- Type
- property vectorListDescriptor
Vector list descriptor memory category.
The data for VectorList[Input[…] DeltaN] fields.
- Type
- property vertexCode
Vertex code memory category.
Code for vertex functions (codelets).
- Type
- property vertexFieldData
Vertex field data memory category.
Variable-sized fields, for example the data for Vector<float>, Vector<Input<…>> and InputSet<…> fields.
- Type
- property vertexInstanceState
Vertex instance state memory category.
An instance of a Vertex class object. This will be sizeof(VertexName) for each vertex.
- Type
- class pva.MemoryWithAndWithoutGaps
MemoryWithAndWithoutGaps represents the amount of memory with and without gaps between allocations.
The memory usage in each region is provided, both with and without gaps. Gaps arise because of memory allocation constraints, such as alignment requirements. For more information on the tile memory architecture, refer to the IPU Programmer’s Guide. Memory with and without gaps between allocations.
- class pva.NotAlwaysLiveMemory
- property bytes
- property variables
List of not always live variables for this step.
- Type
List[pva.Variable]
- class pva.Memory
Memory represents the memory layout of a tile in various ways.
The memory object contains a lot of information about memory use. All memory is statically allocated so you don’t need to run the program to gather this data.
- property alwaysLiveBytes
The total always live memory for this tile.
Variables that exist for the entire execution are always live.
- Type
- property category
Details memory by category.
- Type
- property interleaved
Details of memory with the interleaved region for this tile.
- property nonInterleaved
Details of memory with the non interleaved region for this tile.
- property notAlwaysLiveBytes
The total not always live memory for this tile.
Variables that are created and/or deleted during execution are not. always live.
- Type
- property overflowed
Details of memory with the overflowed region for this tile.
- property total
Details of the total memory usage for this tile.
- property vertices
List of vertex memory usage on this tile.
- Type
List[VertexMemory]
- class pva.Tile
Tile represents details of a single tile on an IPU.
- property memory
Details of memory usage by this tile.
- Type
- class pva.IPU
IPU represents details of a single IPU.
- class Architecture
Enum of the different IPU architectures
Members:
Ipu1 : IPU Mk1
Ipu2 : IPU Mk2
Unknown : Unknown
- property name
- property architecture
The architecture type of this IPU.
- Type
- class pva.IPU.Architecture
Enum of the different IPU architectures
Members:
Ipu1 : IPU Mk1
Ipu2 : IPU Mk2
Unknown : Unknown
- Ipu1 = <Architecture.Ipu1: 0>
- Ipu2 = <Architecture.Ipu2: 1>
- Unknown = <Architecture.Unknown: 4>
- property name
- property value
- class pva.Replica
Replica represents details of a single replica.
- class pva.ComputeSet
ComputeSet represents details of a single compute set.
- property debugContexts
List of Debug Contexts. May be empty.
- Type
List[pva.DebugContext]
- property vars
The lowered variables used by the vertices of this compute set.
- Type
List[pva.LoweredVariable]
- property vertices
The vertices used by this compute set.
- Type
List[pva.VertexInstances]
- class pva.Program
Program is a base class that represents a program type.
- Block = <Type.Block: 22>
- Call = <Type.Call: 19>
- CodeCopy = <Type.CodeCopy: 21>
- DoExchange = <Type.DoExchange: 8>
- GS1 = <SyncType.GS1: 1>
- GS2 = <SyncType.GS2: 2>
- GS3 = <SyncType.GS3: 3>
- GS4 = <SyncType.GS4: 4>
- GetGlobalConsensus = <Type.GetGlobalConsensus: 17>
- GlobalExchange = <Type.GlobalExchange: 9>
- IfElse = <Type.IfElse: 7>
- Internal = <SyncType.Internal: 0>
- OnEveryTileSwitch = <Type.OnEveryTileSwitch: 6>
- OnTileExecute = <Type.OnTileExecute: 2>
- OnTileSwitch = <Type.OnTileSwitch: 5>
- Repeat = <Type.Repeat: 3>
- RepeatWhile = <Type.RepeatWhile: 4>
- Sans = <Type.Sans: 18>
- Sequence = <Type.Sequence: 1>
- SetLocalConsensus = <Type.SetLocalConsensus: 15>
- SetLocalConsensusFromVar = <Type.SetLocalConsensusFromVar: 16>
- StreamCopyBegin = <Type.StreamCopyBegin: 10>
- StreamCopyEnd = <Type.StreamCopyEnd: 12>
- StreamCopyMid = <Type.StreamCopyMid: 11>
- Sync = <Type.Sync: 14>
- SyncAns = <Type.SyncAns: 20>
- class SyncType
Members:
Internal
GS1
GS2
GS3
GS4
Unknown
- GS1 = <SyncType.GS1: 1>
- GS2 = <SyncType.GS2: 2>
- GS3 = <SyncType.GS3: 3>
- GS4 = <SyncType.GS4: 4>
- Internal = <SyncType.Internal: 0>
- Unknown = <SyncType.Unknown: 5>
- property name
- property value
- class Type
Members:
Unknown
Sequence
OnTileExecute
Repeat
RepeatWhile
OnTileSwitch
OnEveryTileSwitch
IfElse
DoExchange
GlobalExchange
StreamCopyBegin
StreamCopyMid
StreamCopyEnd
WriteUndef
Sync
SetLocalConsensus
SetLocalConsensusFromVar
GetGlobalConsensus
Sans
Call
SyncAns
CodeCopy
Block
- Block = <Type.Block: 22>
- Call = <Type.Call: 19>
- CodeCopy = <Type.CodeCopy: 21>
- DoExchange = <Type.DoExchange: 8>
- GetGlobalConsensus = <Type.GetGlobalConsensus: 17>
- GlobalExchange = <Type.GlobalExchange: 9>
- IfElse = <Type.IfElse: 7>
- OnEveryTileSwitch = <Type.OnEveryTileSwitch: 6>
- OnTileExecute = <Type.OnTileExecute: 2>
- OnTileSwitch = <Type.OnTileSwitch: 5>
- Repeat = <Type.Repeat: 3>
- RepeatWhile = <Type.RepeatWhile: 4>
- Sans = <Type.Sans: 18>
- Sequence = <Type.Sequence: 1>
- SetLocalConsensus = <Type.SetLocalConsensus: 15>
- SetLocalConsensusFromVar = <Type.SetLocalConsensusFromVar: 16>
- StreamCopyBegin = <Type.StreamCopyBegin: 10>
- StreamCopyEnd = <Type.StreamCopyEnd: 12>
- StreamCopyMid = <Type.StreamCopyMid: 11>
- Sync = <Type.Sync: 14>
- SyncAns = <Type.SyncAns: 20>
- Unknown = <Type.Unknown: 0>
- WriteUndef = <Type.WriteUndef: 13>
- property name
- property value
- Unknown = <SyncType.Unknown: 5>
- WriteUndef = <Type.WriteUndef: 13>
- accept(self: pva.pva_core.Program, arg0: pva::ProgramVisitor) None
The visitor pattern accept method.
- property children
List of child programs. May be empty.
- Type
List[pva.Program]
- property controlCodeByTile
The size in bytes of the control code of this program and its children (if any) for each tile.
- Type
List[int]
- property debugContexts
List of Debug Contexts. May be empty.
- Type
List[pva.DebugContext]
- property type
The type of the program.
- Type
- property vars
The lowered variables used by this program.
- Type
List[pva.LoweredVariable]
- class pva.OnTileExecuteProgram
- property computeset
The compute set used by this on-tile execute program.
- Type
pva.Computeset
- class pva.UnknownProgram
- class pva.SequenceProgram
- class pva.RepeatProgram
- class pva.RepeatWhileProgram
- class pva.OnTileSwitchProgram
- class pva.OnEveryTileSwitchProgram
- class pva.IfElseProgram
- class pva.DoExchangeProgram
-
- property codeBytesByTile
The size of the code required for this program on each tile.
- Type
List[int]
- property estimatedCyclesByTile
The estimated number of cycles used to execute this program.
- Type
List[int]
- property vars
The lowered variables used by this exchange.
- Type
List[pva.LoweredVariable]
- class pva.GlobalExchangeProgram
-
- property estimatedCyclesByTile
The estimated number of cycles used to execute this program.
- Type
List[int]
- property syncType
The type of synchronisation used by this program.
- Type
- property vars
The lowered variables used by this exchange.
- Type
List[pva.LoweredVariable]
- class pva.StreamCopyBeginProgram
- class pva.StreamCopyMidProgram
- Host = <StreamCopyType.Host: 0>
- Mixed = <StreamCopyType.Mixed: 2>
- RemoteBuffer = <StreamCopyType.RemoteBuffer: 1>
- class StreamCopyType
Indicates whether the stream copy is to/from the host or a remote buffer.
Members:
Host
RemoteBuffer
Mixed
Unknown
- Host = <StreamCopyType.Host: 0>
- Mixed = <StreamCopyType.Mixed: 2>
- RemoteBuffer = <StreamCopyType.RemoteBuffer: 1>
- Unknown = <StreamCopyType.Unknown: 3>
- property name
- property value
- Unknown = <StreamCopyType.Unknown: 3>
- property estimatedCyclesByTile
The estimated number of cycles used to execute this program.
- Type
List[int]
- property streamCopyType
Indicates whether this stream copy is to/from the host or a remote buffer.
- property vars
The lowered variables used by this exchange.
- Type
List[pva.LoweredVariable]
- class pva.StreamCopyEndProgram
- class pva.WriteUndefProgram
- class pva.SetLocalConsensusProgram
- class pva.SetLocalConsensusFromVarProgram
- property children
List of child programs.
- Type
List[pva.Program]
- class pva.GetGlobalConsensusProgram
- class pva.SansProgram
- class pva.SyncAnsProgram
- class pva.ProgramVisitor
- visitCall(self: pva.pva_core.ProgramVisitor, arg0: pva.pva_core.CallProgram) None
- visitCodeCopy(self: pva.pva_core.ProgramVisitor, arg0: pva.pva_core.CodeCopyProgram) None
- visitDoExchange(self: pva.pva_core.ProgramVisitor, arg0: pva.pva_core.DoExchangeProgram) None
- visitGetGlobalConsensus(self: pva.pva_core.ProgramVisitor, arg0: pva.pva_core.GetGlobalConsensusProgram) None
- visitGlobalExchange(self: pva.pva_core.ProgramVisitor, arg0: pva.pva_core.GlobalExchangeProgram) None
- visitIfElse(self: pva.pva_core.ProgramVisitor, arg0: pva.pva_core.IfElseProgram) None
- visitOnEveryTileSwitch(self: pva.pva_core.ProgramVisitor, arg0: pva.pva_core.OnEveryTileSwitchProgram) None
- visitOnTileExecute(self: pva.pva_core.ProgramVisitor, arg0: pva.pva_core.OnTileExecuteProgram) None
- visitOnTileSwitch(self: pva.pva_core.ProgramVisitor, arg0: pva.pva_core.OnTileSwitchProgram) None
- visitProgram(self: pva.pva_core.ProgramVisitor, arg0: pva.pva_core.Program) None
- visitRepeat(self: pva.pva_core.ProgramVisitor, arg0: pva.pva_core.RepeatProgram) None
- visitRepeatWhile(self: pva.pva_core.ProgramVisitor, arg0: pva.pva_core.RepeatWhileProgram) None
- visitSans(self: pva.pva_core.ProgramVisitor, arg0: pva.pva_core.SansProgram) None
- visitSequence(self: pva.pva_core.ProgramVisitor, arg0: pva.pva_core.SequenceProgram) None
- visitSetLocalConsensus(self: pva.pva_core.ProgramVisitor, arg0: pva.pva_core.SetLocalConsensusProgram) None
- visitSetLocalConsensusFromVar(self: pva.pva_core.ProgramVisitor, arg0: pva.pva_core.SetLocalConsensusFromVarProgram) None
- visitStreamCopyBegin(self: pva.pva_core.ProgramVisitor, arg0: pva.pva_core.StreamCopyBeginProgram) None
- visitStreamCopyEnd(self: pva.pva_core.ProgramVisitor, arg0: pva.pva_core.StreamCopyEndProgram) None
- visitStreamCopyMid(self: pva.pva_core.ProgramVisitor, arg0: pva.pva_core.StreamCopyMidProgram) None
- visitSync(self: pva.pva_core.ProgramVisitor, arg0: pva.pva_core.SyncProgram) None
- visitSyncAns(self: pva.pva_core.ProgramVisitor, arg0: pva.pva_core.SyncAnsProgram) None
- visitUnknown(self: pva.pva_core.ProgramVisitor, arg0: pva.pva_core.UnknownProgram) None
- visitWriteUndef(self: pva.pva_core.ProgramVisitor, arg0: pva.pva_core.WriteUndefProgram) None
- class pva.VariableSize
VariableSize represents details of a variable with size information.
- class pva.VariableCategory
This class describes the category of the data stored in a variable (for example Message, Constant, Stack).
- property description
The description of this category.
- Type
string
- property name
The name of this category.
- Type
string
- class pva.Variable
Variable represents details of a variable.
- property debugContexts
List of debug contexts of the variable. May be empty.
- Type
List[pva.DebugContext]
- class pva.LivenessProgramStep
LivenessProgramStep represents details liveness for step in the program.
- notAlwaysLiveBytesForVariable(self: pva.pva_core.LivenessProgramStep, arg0: pva.pva_core.Variable) int
- property notAlwaysLiveMemory
- notAlwaysLiveMemoryForIpu(self: pva.pva_core.LivenessProgramStep, arg0: int) pva.pva_core.NotAlwaysLiveMemory
- notAlwaysLiveMemoryForTile(self: pva.pva_core.LivenessProgramStep, arg0: int) pva.pva_core.NotAlwaysLiveMemory
- property program
- class pva.LoweredVariable
Contains information about a lowered variable.
- property allocated
True if this variable has been allocated to a tile. If not, the rest of the struct should be ignored..
- Type
- property alwaysLive
If true, this variable can never overlap with others. Other variables may end up always live as a result of their usage, but those flagged as such we know should be always live and will be discounted for liveness analysis and reporting.
- Type
- property category
Category for the data stored in the variable.
- Type
- property elementConstraints
The variables that cannot be stored in the same memory element as this variable.
- Type
List[pva.LoweredVariable]
- property equivalenceClass
The equivalence class of this variable. Variables with the same equivalence class have the same liveness profile.
- Type
- property executable
If true, this variable needs to be allocated in an executable region of memory.
- Type
- property lowest256KBytes
If true, this variable contains vertex state that should be in the first 256KB of memory so that it can be addressed via the run instruction.
- Type
- property name
The name of this lowered variable.
- Type
string
- property regionConstraints
The variables that cannot be stored in the same memory region as this variable.
- Type
List[pva.LoweredVariable]
- property startSection
If true, this variable contains an entrypoint that may need special allocation.
- Type
- property unloweredVar
The unlowered variable that originated this variable. Note an exception is thrown if
hasUnloweredVaris false.- Type
- class pva.LoweredVariableBrief
Contains brief information about a lowered variable.
- property equivalenceClass
The liveness equivalence class of the variable.
- Type
- class pva.LoweredVariables
This class allows access to lowered variables by tile and by unlowered variable.
- property allBriefVars
Brief information for each lowered variable.
- Type
List[pva.LoweredVariableBrief]
- forTile(self: pva.pva_core.LoweredVariables, arg0: int) pva::List<pva::LoweredVariable>
List[pva.LoweredVariable]: A list of lowered variables for a given tile.
- forVar(self: pva.pva_core.LoweredVariables, arg0: pva.pva_core.Variable) pva::List<pva::LoweredVariable>
- List[pva.LoweredVariable]: A list of lowered variables for a given
unlowered variable.
- class pva.EquivalenceClass
This class describes the liveness profile of a group of lowered variables.
- property interferenceClasses
The equivalence classes that interfere with this equivalence class.
- Type
List[pva.EquivalenceClass]
- property steps
The list of step intervals where this class is live.
- Type
List[pva.Interval]
- class pva.CompilationReport
Information from the compilation of your model.
- property allocationOrderByTile
For each tile, a list of variable ids in the same order they were allocated in memory.
- Type
List[List[int]]
- property alwaysLiveVariables
List of always live variables.
- Type
List[pva.Variable]
- alwaysLiveVariablesForTile(self: pva.pva_core.CompilationReport, arg0: int) pva.pva_core.VariableSizes
List[pva.Variable]: List of always live variables for a given tile.
- property compilationParameters
A list of all compilation parameters used in poplar
- Type
List[pva.EngineOption]
- property computeSets
A list of all compute sets for all steps.
- Type
List[pva.ComputeSet]
- property controlPrograms
List of all control programs.
- Type
List[pva.Program]
- debugContext(self: pva.pva_core.CompilationReport, arg0: int) pva.pva_core.DebugContext
pva.DebugContext: Get
DebugContextby ID.
- debugContexts(self: pva.pva_core.CompilationReport, arg0: pva.pva_core.DebugContextFilter) pva.pva_core.DebugContexts
- List[pva.DebugContext]: List of debug contexts
filtered by
pva.DebugContextFilter.
- property functions
List of all functions.
- Type
List[pva.Program]
- property livenessProgramSteps
List of program steps with liveness information.
- Type
List[pva.LivenessProgramStep]
- property loweredVariables
An accessor object that contains the lowered variables.
- Type
- property programs
List of all programs.
- Type
List[pva.Program]
- property replicas
List of Replicas.
- Type
List[pva.Replica]
- property target
Information about the target hardware.
- Type
- property targetParameters
A list of all target parameters used in poplar
- Type
List[pva.EngineOption]
- class pva.EngineOption
A Poplar option which controls the behaviour of a graph at compile-time.
- class pva.Variable
Variable represents details of a variable.
- property debugContexts
List of debug contexts of the variable. May be empty.
- Type
List[pva.DebugContext]
- class pva.DebugContext
DebugContext describes the context of an operation (e.g. variable and program creation)
- property children
List of child debug contexts.
- Type
List[pva.DebugContext]
- property location
The location in code where the debug context was created.
- Type
DebugContextLocation
- property parents
The immediate parent debug contexts.
- Type
List[pva.DebugContext]
- programs(self: pva.pva_core.DebugContext, inlineCalls: bool = True) pva::List<std::shared_ptr<pva::Program> >
List[pva.Programs]: List of programs that correspond to this debug context or any of its descendants.
- property variables
The variables that correspond to this debug context or any of its descendants.
- Type
List[pva.Variable]
- class pva.DebugContextFilter
This class describes a filter to be used when fetching debug contexts.
- class pva.TileCycleTotals
This class contains the total cycles for various activities, such as exchange and compute.
3.2. Execution reports
- class pva.Run
Run contains information about each Poplar engine run.
- property cyclesByIpu
The cycles by IPU at the start and end of this run.
Note: The vector may be empty if the run was too short to record cycles.
- Type
VectorList[pva.Period[int]]
- property executionParameters
The execution parameters used for this run.
- Type
List[pva.EngineOption]
- property microseconds
The start and end of this run in microseconds.
- Type
- property steps
A list of program steps in this run.
- Type
List[ExecutionStep]
- class pva.ExecutionStepIpu
- property activeCycles
Statistics for the cycles on active tiles.
Note : Will return zeros if this IPU has not been profiled.
- Type
pva.CyclesInfo
- property activeTileBalance
Indication of active hardware tile utilisation.
Measures how well-utilised the active hardware tiles are. Larger values (up to a maximum of 1.0) indicate that work has been shared amongst the active tiles on this IPU more evenly.
- NoteOnly valid for OnTileExecute, GlobalExchange, StreamCopy,
DoExchange.
Note : Will return 0 if this IPU has not been profiled.
- Type
- property activeTiles
The range of cycles for all tiles involved on this IPU.
Note : Will return 0 if this IPU has not been profiled.
- Type
- property allCycles
Statistics for the cycles on all tiles.
Note : Will return zeros if this IPU has not been profiled.
- Type
pva.CyclesInfo
- property cycles
Total cycles executed on this IPU. Note : will return 0 if this IPU has not been profiled.
- Type
- property dataBalance
Indication of how well-balanced the data transfer was.
Note : Only valid for GlobalExchange, StreamCopy, DoExchange
Note : Will return 0 if this IPU has not been profiled.
- Type
- property dataIn
Total data received by this IPU.
With the Poplar Engine option ‘replicaToProfile’, only a subset of the IPU may have execution profile information.
Note : Only valid for GlobalExchange, StreamCopy, DoExchange. Note : Will return 0 if this IPU has not been profiled.
- Type
- property dataOut
Total data sent by this IPU.
With the Poplar Engine option ‘replicaToProfile’, only a subset of the IPU may have execution profile information.
Note : Only valid for GlobalExchange, StreamCopy, DoExchange. Note : Will return 0 if this IPU has not been profiled.
- Type
- property profiled
Indication if this IPU was profiled.
With the Poplar Engine option ‘replicaToProfile’, only a subset of the IPUs may have execution profile information.
- Type
- property threadBalance
Indication of hardware thread utilisation.
Measures how well-utilised the hardware threads are. If you always run 6 threads or 0 threads this will be 1.0 even if the total computation on each tile takes a different amount of time.
Note : Will return 0 if this IPU has not been profiled.
- Type
- property tileBalance
Indication of hardware tile utilisation.
Measures how well-utilised the hardware tiles are. Larger values (up to a maximum of 1.0) indicate that work has been shared amongst tiles on this IPU more evenly.
- NoteOnly valid for OnTileExecute, GlobalExchange, StreamCopy,
DoExchange.
Note : Will return 0 if this IPU has not been profiled.
- Type
- class pva.ExecutionStep
ExecutionStep contains information about a step of execution. Each step will represent the execution of Poplar program.
- property computeSets
A list of all compute sets for this step.
- Type
List[pva.execution.ComputeSet]
- property cyclesByTile
The measured cycles-by-tile for this execution step.
When using the ‘replicaToProfile’ option, tiles on the replica profiled will have measured values, the other tiles will return 0
- Type
List[int]
- property ipus
Details of a program step execution per IPU.
- Type
List[pva.ExecutionStepIpu]
- property program
The program being executed.
- Type
- class pva.ExecutionStepCycleRange
Details of the min and max cycles.
- class pva.ExecutionReport
ExecutionReportcontains information collected from the execution of a model.- blocks(self: pva.pva_core.ExecutionReport, arg0: int) pva.pva_core.Blocks
List[pva.Block]: A list of all block measurements. May be empty.
- property steps
A list of all program steps for the entire execution.
- Type
List[pva.ExecutionStep]
- property totalCycles
Total cycles executed on all IPUs. For IPUs and replicas.
- Type
- class pva.Period
A period of time measured in microseconds, cycles, or other units.
- class pva.Block
This class represents the cycle measurements around a given program subtree performed by lightweight profiling.
- property debugContext
DebugContextfor this block. An exception is thrown ifCompilationReport::isDebugInfoPresentis false or ifisOverflow()orisAnonymous()orisUnknown()is true.- Type
- property isAnonymous
True if this block has no
DebugContextbecause it began while the buffer was full.- Type
- property isOverflow
True if this block represents a period of time when the instrumentation buffer was full.
- Type
- property program
The program of this block. An exception is thrown if
isOverflow()orisAnonymous()orisUnknown()is true.- Type
3.3. Trace reports
- class pva.TraceReport
TraceReportcontains information collected from the execution of a model on the CPU.- process(self: pva.pva_core.TraceReport, arg0: int) pva.pva_core.Process
pva.Process: Retrieve a process based on its process ID.
- property processes
A list of all processes traced on the CPU.
- Type
List[pva.Process]
- class pva.Process
Information about a process, such as its threads.
- thread(self: pva.pva_core.Process, arg0: int) pva.pva_core.Thread
pva.Thread: Retrieve a thread based on its thread ID.
- property threads
A list of all threads belonging to a process.
- Type
List[pva.Thread]
- class pva.Thread
Information about a thread, such as its events.
- class pva.Event
Information about an event, such as its channel, label, and children.
- children(self: pva.pva_core.Event) List[pva.pva_core.Event]
List[pva.Event]: A list of all events that occurred within this one.