hls4ml.backends.oneapi.passes package

Submodules

hls4ml.backends.oneapi.passes.bn_quant module

class hls4ml.backends.oneapi.passes.bn_quant.BatchNormalizationQuantizedTanhConfigTemplate

Bases: LayerConfigTemplate

format(node)
class hls4ml.backends.oneapi.passes.bn_quant.BatchNormalizationQuantizedTanhFunctionTemplate

Bases: FunctionCallTemplate

format(node)
class hls4ml.backends.oneapi.passes.bn_quant.BatchNormalizationQuantizedTanhStreamFunctionTemplate

Bases: StreamFunctionCallTemplate

format(node)
class hls4ml.backends.oneapi.passes.bn_quant.BatchNormalizationQuantizedTanhTaskSequenceTemplate

Bases: TaskSequenceTemplate

format(node)
class hls4ml.backends.oneapi.passes.bn_quant.MergeBatchNormAndQuantizedTanh

Bases: OptimizerPass

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.

class hls4ml.backends.oneapi.passes.bn_quant.QuantizeDenseOutput

Bases: OptimizerPass

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.backends.oneapi.passes.bn_quant.register_bn_quant(backend)

hls4ml.backends.oneapi.passes.clone_templates module

The clone templates in the fpga backend are not enough for oneAPI, so this adds the missing parts

class hls4ml.backends.oneapi.passes.clone_templates.CloneStreamFunctionTemplate

Bases: StreamFunctionCallTemplate

format(node)
class hls4ml.backends.oneapi.passes.clone_templates.CloneTaskSequenceTemplate

Bases: TaskSequenceTemplate

format(node)

hls4ml.backends.oneapi.passes.convolution_templates module

class hls4ml.backends.oneapi.passes.convolution_templates.Conv1DConfigTemplate

Bases: LayerConfigTemplate

format(node)
class hls4ml.backends.oneapi.passes.convolution_templates.Conv1DFunctionTemplate

Bases: FunctionCallTemplate

format(node)
class hls4ml.backends.oneapi.passes.convolution_templates.Conv1DTaskSequenceTemplate

Bases: TaskSequenceTemplate

format(node)
class hls4ml.backends.oneapi.passes.convolution_templates.Conv2DConfigTemplate

Bases: LayerConfigTemplate

format(node)
class hls4ml.backends.oneapi.passes.convolution_templates.Conv2DFunctionTemplate

Bases: FunctionCallTemplate

format(node)
class hls4ml.backends.oneapi.passes.convolution_templates.Conv2DTaskSequenceTemplate

Bases: TaskSequenceTemplate

format(node)
class hls4ml.backends.oneapi.passes.convolution_templates.ConvStreamFunctionTemplate

Bases: StreamFunctionCallTemplate

format(node)
hls4ml.backends.oneapi.passes.convolution_templates.conv_mult_config_template = 'struct config{index}_mult : nnet::dense_config {{\n    static const unsigned n_in = {n_in};\n    static const unsigned n_out = {n_out};\n\n    static const unsigned rf_pad = {rfpad};\n    static const unsigned bf_pad = {bfpad};\n\n    static const unsigned reuse_factor = {reuse};\n    static const unsigned reuse_factor_rounded = reuse_factor + rf_pad;\n    static const unsigned block_factor = DIV_ROUNDUP(n_in*n_out, reuse_factor);\n    static const unsigned block_factor_rounded = block_factor + bf_pad;\n    static const unsigned multiplier_factor = MIN(n_in, reuse_factor);\n    static const unsigned multiplier_limit = DIV_ROUNDUP(n_in*n_out, multiplier_factor);\n    static const unsigned multiplier_scale = multiplier_limit/n_out;\n\n    typedef {accum_t.name} accum_t;\n    typedef {bias_t.name} bias_t;\n    typedef {weight_t.name} weight_t;\n\n    template<class x_T, class y_T>\n    using product = nnet::product::{product_type}<x_T, y_T>;\n}};\n'

1D Conv

hls4ml.backends.oneapi.passes.convolution_winograd module

class hls4ml.backends.oneapi.passes.convolution_winograd.ApplyWinogradKernelTransformation

Bases: OptimizerPass

Transforms the weights of a Conv2D kernel to a format suitable for Wingorad convolution For further information, refer to Lavin & Gray, 2015 - Fast Algorithms for Convolutional Neural Networks

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.backends.oneapi.passes.core_templates module

class hls4ml.backends.oneapi.passes.core_templates.ActivationConfigTemplate

Bases: LayerConfigTemplate

format(node)
class hls4ml.backends.oneapi.passes.core_templates.ActivationFunctionTemplate

Bases: FunctionCallTemplate

format(node)
class hls4ml.backends.oneapi.passes.core_templates.ActivationStreamFunctionTemplate

Bases: StreamFunctionCallTemplate

format(node)
class hls4ml.backends.oneapi.passes.core_templates.ActivationTaskSequenceTemplate

Bases: TaskSequenceTemplate

format(node)
class hls4ml.backends.oneapi.passes.core_templates.BatchNormalizationConfigTemplate

Bases: LayerConfigTemplate

format(node)
class hls4ml.backends.oneapi.passes.core_templates.BatchNormalizationFunctionTemplate

Bases: FunctionCallTemplate

format(node)
class hls4ml.backends.oneapi.passes.core_templates.BatchNormalizationStreamFunctionTemplate

Bases: StreamFunctionCallTemplate

format(node)
class hls4ml.backends.oneapi.passes.core_templates.BatchNormalizationTaskSequenceTemplate

Bases: TaskSequenceTemplate

format(node)
class hls4ml.backends.oneapi.passes.core_templates.DenseConfigTemplate

Bases: LayerConfigTemplate

format(node)
class hls4ml.backends.oneapi.passes.core_templates.DenseFunctionTemplate

Bases: FunctionCallTemplate

format(node)
class hls4ml.backends.oneapi.passes.core_templates.DenseStreamFunctionTemplate

Bases: StreamFunctionCallTemplate

format(node)
class hls4ml.backends.oneapi.passes.core_templates.DenseTaskSequenceTemplate

Bases: TaskSequenceTemplate

format(node)
class hls4ml.backends.oneapi.passes.core_templates.HardActivationConfigTemplate

Bases: LayerConfigTemplate

format(node)
class hls4ml.backends.oneapi.passes.core_templates.PReLUActivationStreamFunctionTemplate

Bases: StreamFunctionCallTemplate

format(node)
class hls4ml.backends.oneapi.passes.core_templates.PReLUFunctionTemplate

Bases: FunctionCallTemplate

format(node)
class hls4ml.backends.oneapi.passes.core_templates.ParamActivationConfigTemplate

Bases: LayerConfigTemplate

format(node)
class hls4ml.backends.oneapi.passes.core_templates.ParametrizedActivationFunctionTemplate

Bases: FunctionCallTemplate

format(node)
class hls4ml.backends.oneapi.passes.core_templates.ParametrizedActivationStreamFunctionTemplate

Bases: StreamFunctionCallTemplate

format(node)
class hls4ml.backends.oneapi.passes.core_templates.ParametrizedActivationTaskSequenceTemplate

Bases: TaskSequenceTemplate

format(node)
class hls4ml.backends.oneapi.passes.core_templates.SoftmaxConfigTemplate

Bases: ActivationConfigTemplate

hls4ml.backends.oneapi.passes.embedding_templates module

These are the stream oneAPI templates for embedding layers. The io_parallel ones are in backends/fpga/passes/embedding.py.

class hls4ml.backends.oneapi.passes.embedding_templates.EmbeddingStreamFunctionTemplate

Bases: StreamFunctionCallTemplate

format(node)
class hls4ml.backends.oneapi.passes.embedding_templates.EmbeddingTaskSequenceTemplate

Bases: TaskSequenceTemplate

format(node)

hls4ml.backends.oneapi.passes.merge_templates module

class hls4ml.backends.oneapi.passes.merge_templates.ConcatenateConfigTemplate

Bases: LayerConfigTemplate

format(node)
class hls4ml.backends.oneapi.passes.merge_templates.DotConfigTemplate

Bases: LayerConfigTemplate

format(node)
class hls4ml.backends.oneapi.passes.merge_templates.MergeConfigTemplate

Bases: LayerConfigTemplate

format(node)
class hls4ml.backends.oneapi.passes.merge_templates.MergeFunctionTemplate

Bases: FunctionCallTemplate

format(node)
class hls4ml.backends.oneapi.passes.merge_templates.MergeStreamFunctionTemplate

Bases: StreamFunctionCallTemplate

format(node)
class hls4ml.backends.oneapi.passes.merge_templates.MergeTaskSequenceTemplate

Bases: TaskSequenceTemplate

format(node)

hls4ml.backends.oneapi.passes.pointwise module

class hls4ml.backends.oneapi.passes.pointwise.OptimizePointwiseConv

Bases: OptimizerPass

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.

class hls4ml.backends.oneapi.passes.pointwise.PointwiseConv1DConfigTemplate

Bases: Conv1DConfigTemplate

class hls4ml.backends.oneapi.passes.pointwise.PointwiseConv1DFunctionTemplate

Bases: FunctionCallTemplate

format(node)
class hls4ml.backends.oneapi.passes.pointwise.PointwiseConv1DTaskSequenceTemplate

Bases: TaskSequenceTemplate

format(node)
class hls4ml.backends.oneapi.passes.pointwise.PointwiseConv2DConfigTemplate

Bases: Conv2DConfigTemplate

class hls4ml.backends.oneapi.passes.pointwise.PointwiseConv2DFunctionTemplate

Bases: FunctionCallTemplate

format(node)
class hls4ml.backends.oneapi.passes.pointwise.PointwiseConv2DTaskSequenceTemplate

Bases: TaskSequenceTemplate

format(node)
class hls4ml.backends.oneapi.passes.pointwise.PointwiseConvStreamFunctionTemplate

Bases: StreamFunctionCallTemplate

format(node)
hls4ml.backends.oneapi.passes.pointwise.register_pointwise(backend)

hls4ml.backends.oneapi.passes.pooling_templates module

class hls4ml.backends.oneapi.passes.pooling_templates.PoolingConfigTemplate

Bases: LayerConfigTemplate

format(node)
class hls4ml.backends.oneapi.passes.pooling_templates.PoolingFunctionTemplate

Bases: FunctionCallTemplate

format(node)
class hls4ml.backends.oneapi.passes.pooling_templates.PoolingStreamFunctionTemplate

Bases: StreamFunctionCallTemplate

format(node)
class hls4ml.backends.oneapi.passes.pooling_templates.PoolingTaskSequenceTemplate

Bases: TaskSequenceTemplate

format(node)

hls4ml.backends.oneapi.passes.quantization_templates module

class hls4ml.backends.oneapi.passes.quantization_templates.ApplyAlphaConfigTemplate

Bases: LayerConfigTemplate

format(node)
class hls4ml.backends.oneapi.passes.quantization_templates.ApplyAlphaFunctionTemplate

Bases: FunctionCallTemplate

format(node)
class hls4ml.backends.oneapi.passes.quantization_templates.ApplyAlphaStreamFunctionTemplate

Bases: StreamFunctionCallTemplate

format(node)
class hls4ml.backends.oneapi.passes.quantization_templates.ApplyAlphaTaskSequenceTemplate

Bases: TaskSequenceTemplate

format(node)

hls4ml.backends.oneapi.passes.recurrent_templates module

class hls4ml.backends.oneapi.passes.recurrent_templates.GRUConfigTemplate

Bases: LayerConfigTemplate

format(node)
class hls4ml.backends.oneapi.passes.recurrent_templates.GRUFunctionTemplate

Bases: FunctionCallTemplate

format(node)
class hls4ml.backends.oneapi.passes.recurrent_templates.GRUStreamFunctionTemplate

Bases: StreamFunctionCallTemplate

format(node)
class hls4ml.backends.oneapi.passes.recurrent_templates.GRUTaskSequenceTemplate

Bases: TaskSequenceTemplate

format(node)
class hls4ml.backends.oneapi.passes.recurrent_templates.LSTMConfigTemplate

Bases: LayerConfigTemplate

format(node)
class hls4ml.backends.oneapi.passes.recurrent_templates.LSTMFunctionTemplate

Bases: FunctionCallTemplate

format(node)
class hls4ml.backends.oneapi.passes.recurrent_templates.SimpleRNNConfigTemplate

Bases: LayerConfigTemplate

format(node)
class hls4ml.backends.oneapi.passes.recurrent_templates.SimpleRNNFunctionTemplate

Bases: FunctionCallTemplate

format(node)

hls4ml.backends.oneapi.passes.reshaping_templates module

class hls4ml.backends.oneapi.passes.reshaping_templates.ReshapeConfigTemplate

Bases: LayerConfigTemplate

format(node)
class hls4ml.backends.oneapi.passes.reshaping_templates.ReshapeFunctionTemplate

Bases: FunctionCallTemplate

Only used to add the include list

format(node)
class hls4ml.backends.oneapi.passes.reshaping_templates.ReshapeTaskSequenceTemplate

Bases: TaskSequenceTemplate

format(node)
class hls4ml.backends.oneapi.passes.reshaping_templates.ReshapingStreamFunctionTemplate

Bases: StreamFunctionCallTemplate

format(node)
class hls4ml.backends.oneapi.passes.reshaping_templates.ResizeConfigTemplate

Bases: LayerConfigTemplate

format(node)
class hls4ml.backends.oneapi.passes.reshaping_templates.ResizeFunctionTemplate

Bases: FunctionCallTemplate

format(node)
class hls4ml.backends.oneapi.passes.reshaping_templates.ResizeTaskSequenceTemplate

Bases: TaskSequenceTemplate

format(node)
class hls4ml.backends.oneapi.passes.reshaping_templates.TransposeConfigTemplate

Bases: LayerConfigTemplate

format(node)
class hls4ml.backends.oneapi.passes.reshaping_templates.TransposeFunctionTemplate

Bases: FunctionCallTemplate

format(node)
class hls4ml.backends.oneapi.passes.reshaping_templates.TransposeTaskSequenceTemplate

Bases: TaskSequenceTemplate

format(node)
class hls4ml.backends.oneapi.passes.reshaping_templates.ZeroPaddingConfigTemplate

Bases: LayerConfigTemplate

format(node)
class hls4ml.backends.oneapi.passes.reshaping_templates.ZeroPaddingFunctionTemplate

Bases: FunctionCallTemplate

format(node)
class hls4ml.backends.oneapi.passes.reshaping_templates.ZeroPaddingTaskSequenceTemplate

Bases: TaskSequenceTemplate

format(node)

hls4ml.backends.oneapi.passes.resource_strategy module

class hls4ml.backends.oneapi.passes.resource_strategy.ApplyResourceStrategy

Bases: OptimizerPass

Transposes the weights to use the dense_resource matrix multiply routine

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.backends.oneapi.passes.transform_types module

class hls4ml.backends.oneapi.passes.transform_types.TransformTypes

Bases: GlobalOptimizerPass

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.

Module contents