Deeploy.Targets.Neureka.Platform.MemoryNeurekaPlatform

class Deeploy.Targets.Neureka.Platform.MemoryNeurekaPlatform(memoryHierarchy: ~Deeploy.MemoryLevelExtension.MemoryLevels.MemoryHierarchy, defaultTargetMemoryLevel: ~Deeploy.MemoryLevelExtension.MemoryLevels.MemoryLevel, weightMemoryLevel: ~Deeploy.MemoryLevelExtension.MemoryLevels.MemoryLevel | None = None, engines=[NeurekaEngine(name='Neureka', mappings=['RequantizedConv', 'Conv']), PULPClusterEngine(name='PULPCluster', mappings=['Conv', 'RequantizedConv', 'RequantizedGemm', 'Gemm', 'Gelu', 'LayerNormalization', 'MaxPool', 'RequantizediGELU', 'RQIntegerDiv', 'MatMul', 'IntegerMean', 'iSoftmax', 'Softmax', 'ReduceMean', 'ReduceSum', 'RequantShift', 'Add', 'Flatten', 'Gather', 'Mul', 'Pad', 'Relu', 'Reshape', 'Squeeze', 'Transpose', 'Unsqueeze', 'Slice', 'RequantizedAdd', 'Concat', 'iRMSNorm', 'iHardswish', 'RequantizediHardswish', 'Quant', 'Dequant', 'SoftmaxGrad', 'SoftmaxCrossEntropyLoss', 'SoftmaxCrossEntropyLossGrad', 'SGD'])], variableBuffer=<class 'Deeploy.Targets.PULPOpen.Platform.PULPVariableBuffer'>, constantBuffer=<class 'Deeploy.Targets.Neureka.Platform.NeurekaConstantBuffer'>, structBuffer=<class 'Deeploy.Targets.PULPOpen.Platform.PULPStructBuffer'>, transientBuffer=<class 'Deeploy.Targets.PULPOpen.Platform.PULPTransientBuffer'>)

Bases: MemoryPULPPlatform

Methods

__init__(memoryHierarchy: ~Deeploy.MemoryLevelExtension.MemoryLevels.MemoryHierarchy, defaultTargetMemoryLevel: ~Deeploy.MemoryLevelExtension.MemoryLevels.MemoryLevel, weightMemoryLevel: ~Deeploy.MemoryLevelExtension.MemoryLevels.MemoryLevel | None = None, engines=[NeurekaEngine(name='Neureka', mappings=['RequantizedConv', 'Conv']), PULPClusterEngine(name='PULPCluster', mappings=['Conv', 'RequantizedConv', 'RequantizedGemm', 'Gemm', 'Gelu', 'LayerNormalization', 'MaxPool', 'RequantizediGELU', 'RQIntegerDiv', 'MatMul', 'IntegerMean', 'iSoftmax', 'Softmax', 'ReduceMean', 'ReduceSum', 'RequantShift', 'Add', 'Flatten', 'Gather', 'Mul', 'Pad', 'Relu', 'Reshape', 'Squeeze', 'Transpose', 'Unsqueeze', 'Slice', 'RequantizedAdd', 'Concat', 'iRMSNorm', 'iHardswish', 'RequantizediHardswish', 'Quant', 'Dequant', 'SoftmaxGrad', 'SoftmaxCrossEntropyLoss', 'SoftmaxCrossEntropyLossGrad', 'SGD'])], variableBuffer=<class 'Deeploy.Targets.PULPOpen.Platform.PULPVariableBuffer'>, constantBuffer=<class 'Deeploy.Targets.Neureka.Platform.NeurekaConstantBuffer'>, structBuffer=<class 'Deeploy.Targets.PULPOpen.Platform.PULPStructBuffer'>, transientBuffer=<class 'Deeploy.Targets.PULPOpen.Platform.PULPTransientBuffer'>) → None

Initializes a new deployment platform

Parameters:

  • engines (List[DeploymentEngine]) – List of all available non-host engines

  • variableBuffer (Type[VariableBuffer]) – VariableBuffer subclass with correctly set allocation and deallocation templates

  • constantBuffer (Type[ConstantBuffer]) – ConstantBuffer subclass with correctly set allocation and deallocation templates

  • structBuffer (Type[StructBuffer]) – StructBuffer subclass with correctly set allocation and deallocation templates

  • transientBuffer (Type[TransientBuffer]) – TransientBuffer subclass with correctly set allocation and deallocation templates

__init__(memoryHierarchy, ...[, ...])

Initializes a new deployment platform

getTargetMemoryLevel(node, tensorName, ctxt)

Attributes

untiledOps

engines

A list of all available non-host engines
engines

A list of all available non-host engines

Type:

List[DeploymentEngine]