chop.passes.graph

Contents

chop.passes.graph#

Summary of MaseGraph Analysis Passes#

MaseGraph graph analysis passes#
Pass Name	Usage Example	Summary
`init_metadata_analysis_pass()`	test_add_common_metadata	Initialize each node with the MaseMetadata, this needs to run first before adding any metadata
`add_common_metadata_analysis_pass()`	test_add_common_metadata	Add metadata used for both software and hardware, this needs to run first before calling to add_software_metadata or add_hardware_metadata
`add_software_metadata_analysis_pass()`	test_add_software_metadata	Add hardware-specific metadata, such as which hardware IPs to map and hardware parameters. Details see the pass page
`add_hardware_metadata_analysis_pass()`	test_add_hardware_metadata	Add software-specific metadata, such as sparsity. Details see the pass page
`report_graph_analysis_pass()`	test_report_graph	Generates a report for the graph analysis and prints out an over the model in a table.
`report_node_hardware_type_analysis_pass()`	test_report_node_hardware_type	Generate a report on the hardware type of the nodes in the graph.
`report_node_meta_param_analysis_pass()`	test_report_node_meta_param	Generate a report on the meta parameters of the nodes in the graph.
`report_node_shape_analysis_pass()`	test_report_node_shape	Generate a report on the shape of the nodes in the graph.
`report_node_type_analysis_pass()`	test_report_node_type	Generate a report on the type of the nodes in the graph.
`profile_statistics_analysis_pass()`	test_profile_statistics	Perform profile statistics analysis on the given graph
`calculate_avg_bits_mg_analysis_pass()`	calculate_avg_bits_mg_analysis_pass	Calculate, on average, how many bits are spent on weights and activations
`add_natural_sparsity_metadata_analysis_pass()`	test_calculate_natural_sparsity	Add natural sparsity metadata analysis pass to the given MaseGraph.
`add_pruning_metadata_analysis_pass()`	test_prune	This pass computes weight and activation sparsity based on pruning masks
`hook_inspection_analysis_pass()`	test_hook_inspection_analysis_pass	Provide hook information of the modules
`runtime_analysis_pass()`	fill me	Perform runtime analysis on the given graph (MaseGraph, TensorRT, ONNX models)
`verify_metadata_analysis_pass()`	fill me	fill me
`verify_common_metadata_analysis_pass()`	fill me	fill me
`verify_software_metadata_analysis_pass()`	fill me	fill me
`verify_hardware_metadata_analysis_pass()`	fill me	fill me

Full list of graph-level analysis passes

Summary of MaseGraph Transform Passes#

MaseGraph graph transform passes#
Pass Name	Usage Example	Summary
`prune_transform_pass()`	test_prune	Prune the given graph
`prune_detach_hook_transform_pass()`	test_prune_detach_hook	Remove all pruning hooks
`quantize_transform_pass()`	test_quantize	Apply quantization, check the further documentation below
`summarize_quantization_analysis_pass()`	test_quantize	Summarizes the quantization with respect to the original graph
`conv_bn_fusion_transform_pass()`	test_conv_bn_fusion	Perform Conv-BN fusion on the given graph
`logicnets_fusion_transform_pass()`	test_logicnets_fusion (DEV, Disabled)	Perform LogicNets fusion on the given graph (DEV, Disabled)
`onnx_annotate_transform_pass()`	NA (DEV, Disabled)	Convert MaseGraph to ONNX and annotate the relevant layers with sparsity information (DEV, Disabled)
`emit_verilog_top_transform_pass()`	test_emit_verilog_linear	Emit the top-level model design in Verilog.
`emit_verilog_tb_transform_pass()`	Emit test bench and related files for simulation.	test_emit_verilog_linear
`emit_bram_transform_pass()`	Enumerate input parameters of the node and emit a ROM block with handshake interface for each parameter	test_emit_verilog_linear
`emit_mlir_hls_transform_pass()`	Emit the hardware components that generated from MLIR-HLS tools.	test_emit_verilog_linear
`emit_internal_rtl_transform_pass()`	Emit the hardware components that pre-defined in the mase internal library.	test_emit_verilog_linear
`tensorrt_fake_quantize_transform_pass()`	Apply TensorRT fake quantization to the given graph for INT8 quantization calibration	fille me
`tensorrt_calibrate_transform_pass()`	Apply TensorRT calibration to the given graph for INT8 quantization	fille me
`tensorrt_fine_tune_transform_pass()`	Apply TensorRT fine tune to the given graph for quantization aware training	fille me

Summary of MaseGraph Interface Passes#

A summary of all MaseGraph interface passes#
Pass Name	Usage Example	Summary
`load_mase_graph_interface_pass()`	fill me	fill me
`save_mase_graph_interface_pass()`	fill me	fill me
`save_node_meta_param_interface_pass()`	fill me	fill me
`load_node_meta_param_interface_pass()`	fill me	fill me
`tensorrt_engine_interface_pass()`	Converts the given graph to a TensorRT engine model	fill me
`onnx_runtime_interface_pass()`	Converts the given graph to a ONNXRuntime model	fill me

Full list of graph-level interface passes