hls4ml.model.optimizer package

Subpackages

• hls4ml.model.optimizer.passes package
  • Submodules
  • hls4ml.model.optimizer.passes.batchnorm_opt module
    • BatchNormOnnxConstantParameters
      • BatchNormOnnxConstantParameters.match()
      • BatchNormOnnxConstantParameters.transform()
    • ConstantBatchNormFusion
      • ConstantBatchNormFusion.match()
      • ConstantBatchNormFusion.transform()
    • FuseConsecutiveBatchNormalization
      • FuseConsecutiveBatchNormalization.match()
      • FuseConsecutiveBatchNormalization.transform()
    • RemoveNopBatchNormalization
      • RemoveNopBatchNormalization.match()
      • RemoveNopBatchNormalization.transform()
  • hls4ml.model.optimizer.passes.bn_fuse module
    • FuseBatchNormalization
      • FuseBatchNormalization.match()
      • FuseBatchNormalization.transform()
  • hls4ml.model.optimizer.passes.conv_to_convxd module
    • ConvToConvXD
      • ConvToConvXD.match()
      • ConvToConvXD.transform()
  • hls4ml.model.optimizer.passes.conv_to_depthwiseconvxd module
    • ConvToDepthwiseConvXD
      • ConvToDepthwiseConvXD.match()
      • ConvToDepthwiseConvXD.transform()
  • hls4ml.model.optimizer.passes.convert_to_channels_last module
    • ChannelsLastConverter
      • ChannelsLastConverter.match()
      • ChannelsLastConverter.transform()
    • RemoveTransposeBeforeFlatten
      • RemoveTransposeBeforeFlatten.match()
      • RemoveTransposeBeforeFlatten.transform()
  • hls4ml.model.optimizer.passes.expand_layer_group module
    • ExpandLayerGroup
      • ExpandLayerGroup.match()
      • ExpandLayerGroup.transform()
  • hls4ml.model.optimizer.passes.fuse_biasadd module
    • FuseBiasAdd
      • FuseBiasAdd.match()
      • FuseBiasAdd.transform()
  • hls4ml.model.optimizer.passes.hgq_proxy_model module
    • EnforceProxyModelEmbeddedConfig
      • EnforceProxyModelEmbeddedConfig.match()
      • EnforceProxyModelEmbeddedConfig.transform()
    • FixedPointQuantizer
      • FixedPointQuantizer.initialize()
    • UnaryLUT
      • UnaryLUT.initialize()
    • register_hgq_proxy_model()
    • to_hls4ml_fixed()
    • userconf_ifdef()
  • hls4ml.model.optimizer.passes.infer_precision module
    • InferPrecisionTypes
      • InferPrecisionTypes.match()
      • InferPrecisionTypes.transform()
  • hls4ml.model.optimizer.passes.linear module
    • EliminateLinearActivation
      • EliminateLinearActivation.match()
      • EliminateLinearActivation.transform()
    • MergeLinearActivation
      • MergeLinearActivation.match()
      • MergeLinearActivation.transform()
  • hls4ml.model.optimizer.passes.matmul_const_to_dense module
    • MatmulConstToDense
      • MatmulConstToDense.match()
      • MatmulConstToDense.transform()
  • hls4ml.model.optimizer.passes.merge_const module
    • MergeToApplyAlpha
      • MergeToApplyAlpha.match()
      • MergeToApplyAlpha.transform()
    • MergeToApplyAlphaDiv
      • MergeToApplyAlphaDiv.match()
      • MergeToApplyAlphaDiv.transform()
    • MergeTwoConstants
      • MergeTwoConstants.match()
      • MergeTwoConstants.transform()
  • hls4ml.model.optimizer.passes.move_scales module
    • BiasDownAdd
      • BiasDownAdd.match()
      • BiasDownAdd.transform()
    • ScaleDownAdd
      • ScaleDownAdd.match()
      • ScaleDownAdd.transform()
    • ScaleDownConv
      • ScaleDownConv.match()
      • ScaleDownConv.transform()
    • ScaleDownMatMul
      • ScaleDownMatMul.match()
      • ScaleDownMatMul.transform()
  • hls4ml.model.optimizer.passes.multi_dense module
    • ReplaceMultidimensionalDenseWithConv
      • ReplaceMultidimensionalDenseWithConv.match()
      • ReplaceMultidimensionalDenseWithConv.transform()
  • hls4ml.model.optimizer.passes.qkeras module
    • ExtractTernaryThreshold
      • ExtractTernaryThreshold.match()
      • ExtractTernaryThreshold.transform()
    • OutputRoundingSaturationMode
      • OutputRoundingSaturationMode.match()
      • OutputRoundingSaturationMode.precision_string_modify()
      • OutputRoundingSaturationMode.transform()
    • QKerasFactorizeAlpha
      • QKerasFactorizeAlpha.match()
      • QKerasFactorizeAlpha.transform()
    • register_qkeras()
  • hls4ml.model.optimizer.passes.quant_opt module
    • ConstQuantToConstAlpha
      • ConstQuantToConstAlpha.match()
      • ConstQuantToConstAlpha.transform()
    • FuseQuantWithConstant
      • FuseQuantWithConstant.match()
      • FuseQuantWithConstant.transform()
    • QuantConstantParameters
      • QuantConstantParameters.match()
      • QuantConstantParameters.transform()
    • QuantToActivation
      • QuantToActivation.match()
      • QuantToActivation.transform()
    • QuantToAlphaActivationAlpha
      • QuantToAlphaActivationAlpha.match()
      • QuantToAlphaActivationAlpha.transform()
  • hls4ml.model.optimizer.passes.reshape_const module
    • ReshapeConstant
      • ReshapeConstant.match()
      • ReshapeConstant.transform()
  • hls4ml.model.optimizer.passes.resize_remove_constants module
    • ResizeRemoveConstants
      • ResizeRemoveConstants.match()
      • ResizeRemoveConstants.transform()
  • hls4ml.model.optimizer.passes.seperable_to_dw_conv module
    • SeparableToDepthwiseAndConv
      • SeparableToDepthwiseAndConv.match()
      • SeparableToDepthwiseAndConv.transform()
  • hls4ml.model.optimizer.passes.stamp module
    • MakeStamp
      • MakeStamp.transform()
  • hls4ml.model.optimizer.passes.transpose_opt module
    • RemoveNopTranspose
      • RemoveNopTranspose.match()
      • RemoveNopTranspose.transform()
    • RemoveSingleChannelTranspose
      • RemoveSingleChannelTranspose.match()
      • RemoveSingleChannelTranspose.transform()
  • Module contents

Submodules

hls4ml.model.optimizer.optimizer module

class hls4ml.model.optimizer.optimizer.ConfigurableOptimizerPass

Bases: OptimizerPass

An optimizer that can be configured.

Existing instances of this class in the registry can be configured with the configure() method. Multiple instances with different configuration can co-exist if registered with different names.

configure(**kwargs)

get_config()

class hls4ml.model.optimizer.optimizer.GlobalOptimizerPass

Bases: OptimizerPass

Global optimizer that matches on every node in the model graph.

match(node)

Predicate to match on a given node.

Parameters:

node (Layer) – Node in the model graph to try matching the optimizer on.

class hls4ml.model.optimizer.optimizer.LayerOptimizerPass(name, layer_class, transform)

Bases: WrappedOptimizerPass

An wrapper optimizer specific to a layer class.

Commonly used by backends to add extra initialization to a layer instance.

class hls4ml.model.optimizer.optimizer.ModelOptimizerPass(name, transform)

Bases: OptimizerPass

A special optimizer that works with the model itself.

Examples include writing the model to C++/HLS.

transform(model)

Transformation to apply if matching was successful.

Transform should return a boolean value indicating if the model graph was altered (by adding/removing nodes).

Parameters:

• model (ModelGraph) – Model to optimize

• node (Layer) – The matched node in the model graph.

class hls4ml.model.optimizer.optimizer.OptimizerPass

Bases: object

Base optimizer class from which all other optimizer types are derived.

classmethod get_name()

match(node)

Predicate to match on a given node.

Parameters:

node (Layer) – Node in the model graph to try matching the optimizer on.

name = None

transform(model, node)

Transformation to apply if matching was successful.

Transform should return a boolean value indicating if the model graph was altered (by adding/removing nodes).

Parameters:

• model (ModelGraph) – Model to optimize

• node (Layer) – The matched node in the model graph.

class hls4ml.model.optimizer.optimizer.WrappedOptimizerPass(name, condition, transform)

Bases: OptimizerPass

An optimizer class created by wrapping a function call.

Users should generally not create any wrapped optimizer passes manually.

get_name()

match(node)

Predicate to match on a given node.

Parameters:

node (Layer) – Node in the model graph to try matching the optimizer on.

transform(model, node)

Transformation to apply if matching was successful.

Transform should return a boolean value indicating if the model graph was altered (by adding/removing nodes).

Parameters:

• model (ModelGraph) – Model to optimize

• node (Layer) – The matched node in the model graph.

hls4ml.model.optimizer.optimizer.extract_optimizers_from_object(clazz)

hls4ml.model.optimizer.optimizer.extract_optimizers_from_path(opt_path, module_path, initializer=None)

hls4ml.model.optimizer.optimizer.get_available_passes()

Return the list of all registered optimizer passes.

Returns:

List of registered passes.

Return type:

list

hls4ml.model.optimizer.optimizer.get_backend_passes(backend)

Returns the list of optimizer passes belonging to a backend

Parameters:

backend (str) – Name of the backend.

Returns:

List of optimizer names registered with the given backend.

Return type:

list

hls4ml.model.optimizer.optimizer.get_optimizer(name)

Return the optimizer instance registered with the given name.

Parameters:

name (str) – Name of the optimizer in the registry.

Raises:

Exception – If the optimizer with the given name is not found in the registry.

Returns:

The optimizer from the registry.

Return type:

OptimizerPass

hls4ml.model.optimizer.optimizer.layer_optimizer(layer)

Decorator to turn a function into the optimization pass.

Example:

@layer_optimizer(MyLayer)
def init_mylayer(self, layer):
    layer.set_attr('new_attribute', 'some_value')

Parameters:

layer (_type_) – _description_

hls4ml.model.optimizer.optimizer.model_optimizer()

Decorator to turn a function into a model optimizer.

hls4ml.model.optimizer.optimizer.optimize_model(model, passes)

Optimize a given model with the given passes.

The passes are attempted until all passes no longer match or no changes to the model graph occur.

Parameters:

• model (ModelGraph) – The model to optimize.

• passes (list) – List of passes to apply.

Returns:

The set of applied passes (the passes that matched the predicate).

Return type:

set

hls4ml.model.optimizer.optimizer.optimizer_pass(condition)

hls4ml.model.optimizer.optimizer.register_pass(name, opt_cls, backend=None)

Register a new optimizer pass.

Parameters:

• name (str) – Name of the optimizer

• opt_cls (class) – The class of the optimizer.

• backend (str, optional) – Optional backend to register the optimizer to. If not None, the name of the backend will be appended to the name of the registered flow. Defaults to None.

Raises:

Exception – If the optimization pass has already been registered with the given name.

Returns:

The name of the registered optimizer.

Return type:

str

Module contents