Analysis

If you want to know more details about analysis passes, please take a look at section “Analysis Pass” in chapter Internals.

Submodules

Analysis - fpgadataflow
- Analysis Passes (fpgadataflow)

Analysis Passes

qonnx.analysis.base

How to write an analysis pass for QONNX

An analysis pass traverses the graph structure and produces information about certain properties. The convention is to take in a ModelWrapper, and return a dictionary of named properties that the analysis extracts.

qonnx.analysis.inference_cost

qonnx.analysis.inference_cost.aggregate_dict_keys(res_dict)

qonnx.analysis.inference_cost.get_node_tensor_dtypes(model, node)

qonnx.analysis.inference_cost.get_node_tensor_shapes(model, node)

qonnx.analysis.inference_cost.get_node_weight_density(model, w_name)

qonnx.analysis.inference_cost.inference_cost(model, discount_sparsity=True, cost_breakdown=False): Ensure all nodes have unique names prior to calling this analysis pass.

qonnx.analysis.inference_cost.inference_cost_conv(model, node, discount_sparsity)

qonnx.analysis.inference_cost.inference_cost_matmul(model, node, discount_sparsity)

qonnx.analysis.inference_cost.inference_cost_upsample(model, node, discount_sparsity)

qonnx.analysis.tensor_stats

qonnx.analysis.tensor_stats.main()

qonnx.analysis.tensor_stats.tensor_stats(modelwrapper_or_filename, act_dump_dir: str, output_stats_dir: str, *, axes='1', plot=True, report_mode: functools.partial(<function _use_class at 0x7f8655d4c940>, <class 'clize.parser.use_mixin.<locals>._PosWithMixin'>, <class 'clize.parser.use_mixin.<locals>._VarargsWithMixin'>, <class 'clize.parser.use_mixin.<locals>._NamedWithMixin'>, <function unimplemented_parameter at 0x7f8655d4c4c0>, {'case_sensitive': None, 'list_name': 'list', 'values': [('range', ['range', ]'Report ranges'), ('stuck_channel', ['stuck_channel', ]'Report stuck channels'), ('zerostuck_channel', ['zerostuck_channel', ]'Report 0-stuck channels')]}) = 'range')

qonnx.analysis.tensor_stats.update_tensor_stats(tensor, axes, ret_dict={})

qonnx.analysis.topology

qonnx.analysis.topology.all_tensors_f32(model)

Checks whether all tensors have a float32 dtype, extra quantization annotations notwithstanding.

Returns {“all_tensors_f32”: Bool}.

qonnx.analysis.topology.get_per_tensor_fanouts(model): Returns a dictionary of {tensor_name: tensor_fanout} for the model.

qonnx.analysis.topology.is_linear(model)

Checks whether the given model graph is linear. This is done by looking at the fan-out of each tensor. All tensors have a fan-out <= 1 in a linear graph.

Returns {“is_linear”: Bool}.

qonnx.analysis.topology.node_inputs_in_expected_order(model)

Verifies that the node inputs are ordered in the way that QONNX expects them. When a node has a mixture of static (= constant, initialized) inputs and dynamic inputs, the dynamic input should come first, followed by the static one. Only verifiable for a small subset of op_types for now.

Returns {“node_inputs_in_expected_order”: Bool}.

qonnx.analysis.topology.nodes_topologically_sorted(model)

Verifies that graph.node is topologically sorted. This is required by the ONNX specification.

Returns {“nodes_topologically_sorted”: Bool}.

finn.analysis.verify_custom_nodes

finn.analysis.verify_custom_nodes.verify_nodes(model)

Checks if custom ops in graph are correctly built, with all attributes and inputs. Please note that many FINN CustomOps don’t yet implement the verify_node function required for this analysis pass to work correctly.

Returns {node op_type : info_messages}

info_messages: is list of strings about the result of the verification.