Sample RISC-V Zephyr Tracing Report¶

This section contains a sample complete report with Zephyr tracing section on RISC-V platform, generated by Kenning using Zephelin.

Testing configuration:

Environment: Github Actions
Task: Gesture recognition on Magic Wand dataset
Tracing tool: Zephelin
Compiler framework: TVM
Platform: SiFive HiFive Unmatched

Sample RISC-V Zephyr Tracing Report¶

General information for results.json¶

Model framework:

torch ver. 2.13.0+cu130

Used commands

This section was generated using:

python -m kenning.__init__ \
    optimize \
    test \
    report \
    --cfg \
        /home/runner/work/kenning/kenning/scripts/configs/zephyr-tvm-magic-wand-hifive-unmatched-report.yml \
    --measurements \
        ./results.json \
    --report-path \
        /home/runner/work/kenning/kenning/docs/source/generated/sample-riscv-zephyr-tracing-report.md \
    --report-types \
        classification \
        performance \
        renode_stats \
        zephyr_traces \
        model \
    --root-dir \
        /home/runner/work/kenning/kenning/docs/source/ \
    --img-dir \
        /home/runner/work/kenning/kenning/docs/source/generated/img/ \
    --report-name \
        Sample RISC-V Zephyr Tracing Report \
    --verbosity \
        INFO \
    --to-html

Input scenario

{
    "dataset": {
        "type": "kenning.datasets.magic_wand_dataset.MagicWandDataset",
        "parameters": {
            "window_size": 128,
            "window_shift": 128,
            "noise_level": 20,
            "dataset_root": "build/data",
            "inference_batch_size": 1,
            "download_dataset": true,
            "force_download_dataset": false,
            "external_calibration_dataset": null,
            "split_fraction_test": 0.2,
            "split_fraction_val": null,
            "split_seed": 1234,
            "reduce_dataset": 1.0
        }
    },
    "dataconverter": {
        "type": "kenning.dataconverters.modelwrapper_dataconverter.ModelWrapperDataConverter",
        "parameters": {}
    },
    "optimizers": [
        {
            "type": "kenning.optimizers.tvm.TVMCompiler",
            "parameters": {
                "model_framework": "any",
                "target": null,
                "target_attrs": "",
                "target_microtvm_board": "hifive_unmatched/fu740/u74",
                "target_host": null,
                "zephyr_header_template": "gh://antmicro:kenning-zephyr-runtime/lib/kenning_inference_lib/runtimes/tvm/generated/model_impl.h.template;branch=main",
                "zephyr_llext_source_template": null,
                "opt_level": 3,
                "libdarknet_path": "/usr/local/lib/libdarknet.so",
                "compile_use_vm": false,
                "output_conversion_function": "default",
                "conv2d_data_layout": "",
                "conv2d_kernel_layout": "",
                "use_fp16_precision": false,
                "use_int8_precision": false,
                "int8_calibrate_chunk_by": -1,
                "use_tensorrt": false,
                "dataset_percentage": 0.25,
                "module_name": null,
                "compiled_model_path": "./build/compiled-model-magic-wand.graph_data",
                "location": "host"
            }
        }
    ],
    "platform": {
        "type": "kenning.platforms.zephyr.ZephyrPlatform",
        "parameters": {
            "zephyr_build_path": "build/hifive_unmatched",
            "llext_binary_path": null,
            "sensors": null,
            "sensors_frequency": null,
            "enable_zephelin_gdb": false,
            "enable_zephelin": true,
            "zephyr_base": null,
            "uart_port": "/tmp/renode_uart_zthc9uoq/uart",
            "uart_baudrate": 115200,
            "uart_log_port": "/tmp/renode_uart_zthc9uoq/uart_log",
            "uart_log_baudrate": 115200,
            "auto_flash": false,
            "openocd_path": "openocd",
            "sensor": null,
            "number_of_batches": 16,
            "simulated": true,
            "runtime_binary_path": null,
            "platform_resc_path": null,
            "resc_dependencies": [],
            "post_start_commands": [
                "logLevel 3 sysbus.uart1"
            ],
            "disable_opcode_counters": false,
            "disable_profiler": false,
            "profiler_dump_path": "/tmp/renode_profiler_k38wa81h.dump",
            "profiler_interval_step": 10.0,
            "runtime_init_log_msg": "Inference server started",
            "runtime_init_timeout": 30,
            "gdb_port": 3333,
            "renode_log_lines_single_read_limit": 5000,
            "name": "hifive_unmatched/fu740/u74",
            "platforms_definitions": [
                "kenning:///platforms/platforms.yml",
                "/home/runner/work/kenning/kenning/kenning/resources/platforms/platforms.yml"
            ]
        }
    },
    "protocol": {
        "type": "kenning.protocols.uart.UARTProtocol",
        "parameters": {
            "port": "/tmp/renode_uart_zthc9uoq/uart",
            "baudrate": 115200,
            "error_recovery": true,
            "timeout": 30
        }
    },
    "model_wrapper": {
        "type": "kenning.modelwrappers.classification.pytorch_magic_wand.PyTorchMagicWandModelWrapper",
        "parameters": {
            "batch_size": null,
            "learning_rate": null,
            "num_epochs": null,
            "window_size": 128,
            "logdir": null,
            "export_dict": false,
            "model_path": "kenning:///models/classification/magic_wand.pth",
            "model_name": null
        }
    },
    "runtime": {
        "type": "kenning.runtimes.tvm.TVMRuntime",
        "parameters": {
            "save_model_path": "./build/compiled-model-magic-wand.graph_data",
            "target_device_context": "cpu",
            "target_device_context_id": 0,
            "runtime_use_vm": false,
            "llext_binary_path": null,
            "batch_size": 1,
            "disable_performance_measurements": false
        }
    },
    "runtime_builder": {
        "type": "kenning.runtimebuilders.zephyr.ZephyrRuntimeBuilder",
        "parameters": {
            "board": "hifive_unmatched/fu740/u74",
            "application_dir": "/home/runner/work/kenning/kenning/zephyr-workspace/kenning-zephyr-runtime/app",
            "build_dir": "/home/runner/work/kenning/kenning/zephyr-workspace/kenning-zephyr-runtime/build",
            "venv_dir": "/home/runner/work/kenning/kenning/zephyr-workspace/kenning-zephyr-runtime/.west-venv",
            "extra_targets": [
                "board-repl"
            ],
            "extra_build_args": [
                "-DCONFIG_KENNING_TVM_MODEL_PRE_GEN=y"
            ],
            "use_llext": false,
            "run_west_update": false,
            "workspace": "/home/runner/work/kenning/kenning/zephyr-workspace/kenning-zephyr-runtime",
            "output_path": "build/hifive_unmatched",
            "model_framework": "tvm"
        }
    }
}

Inference quality metrics for results.json¶

Bokeh Plot

Figure 20 Confusion matrix¶

Table 10 Inference quality metrics¶
Statistic	Value
Accuracy	0.992857
Mean precision	0.994681
Mean sensitivity	0.994681
G-mean	0.994638

Inference performance metrics for results.json¶

Inference time¶

Bokeh Plot

Figure 21 Inference time¶

Bokeh Plot

Figure 22 Preprocessing and postprocessing time¶

Table 11 Inference time metrics¶
Statistic	Time [s]
First inference duration	0.009222
Mean	0.009712
Median	0.009747
Standard deviation	0.000315
Minimum	0.009006
Maximum	0.010400

Renode performance measurements for results.json¶

Count of instructions used during inference¶

Bokeh Plot

Figure 23 Histogram of used instructions during inference¶

Executed instructions counters¶

Bokeh Application

Figure 24 Count of executed instructions per second for cpu0 during benchmark¶

Bokeh Plot

Figure 25 Cumulative count of executed instructions for cpu0 during benchmark¶

Bokeh Application

Figure 26 Count of executed instructions per second for cpu1 during benchmark¶

Bokeh Plot

Figure 27 Cumulative count of executed instructions for cpu1 during benchmark¶

Bokeh Application

Figure 28 Count of executed instructions per second for cpu2 during benchmark¶

Bokeh Plot

Figure 29 Cumulative count of executed instructions for cpu2 during benchmark¶

Bokeh Application

Figure 30 Count of executed instructions per second for cpu3 during benchmark¶

Bokeh Plot

Figure 31 Cumulative count of executed instructions for cpu3 during benchmark¶

Peripheral access counters¶

Bokeh Application

Figure 32 Count of clint reads per second during benchmark¶

Bokeh Plot

Figure 33 Cumulative count of clint reads during benchmark¶

Bokeh Application

Figure 34 Count of clint writes per second during benchmark¶

Bokeh Plot

Figure 35 Cumulative count of clint writes during benchmark¶

Bokeh Application

Figure 36 Count of uart0 reads per second during benchmark¶

Bokeh Plot

Figure 37 Cumulative count of uart0 reads during benchmark¶

Bokeh Application

Figure 38 Count of uart0 writes per second during benchmark¶

Bokeh Plot

Figure 39 Cumulative count of uart0 writes during benchmark¶

Bokeh Application

Figure 40 Count of uart1 reads per second during benchmark¶

Bokeh Plot

Figure 41 Cumulative count of uart1 reads during benchmark¶

Bokeh Application

Figure 42 Count of uart1 writes per second during benchmark¶

Bokeh Plot

Figure 43 Cumulative count of uart1 writes during benchmark¶

Exceptions counters¶

Bokeh Application

Figure 44 Count of raised exceptions per second during benchmark¶

Bokeh Plot

Figure 45 Cumulative count of raised exceptions during benchmark¶

Instructions stats¶

Instructions counters per inference pass: 1420285
Top 10 instructions and counters per inference pass:
- c.addi: 212017
- flw: 160863
- bne: 95863
- fsw: 76555
- fadd.s: 66052
- addi: 62444
- fmul.s: 62272
- c.mv: 55975
- c.flw: 37731
- ld: 36721

Memory allocation stats¶

Total allocated: 3091632
Total freed: 3033712
Peak allocated: 72256
Compiled model size: 22486.0

Host memory refers to memory of the CPU controlling the accelerator, while device memory is the memory of the accelerator.

Zephyr traces¶

(open in a new tab)

Input model specification for results.json¶

Model visualization¶

Basic model information¶

Table 12 Basic model information¶
Model name	Layer count	Number of parameters	ONNX-converted model size (in bytes)	Model Framework (version)
Main	14	4572	18288	torch ()

Layers¶

Bokeh Plot

Figure 46 Layer operation type counts¶

Table 13 Main model layers¶
Layer #	Name	Operation type	Parameters types	Parameters count	Parameters bytes
1	input_1	Input	FLOAT	0	0
2	/0/Conv	Conv	FLOAT	104	416
3	/2/Relu	Relu	N/A	0	0
4	/3/MaxPool	MaxPool	N/A	0	0
5	/5/Conv	Conv	FLOAT	528	2112
6	/7/Relu	Relu	N/A	0	0
7	/8/MaxPool	MaxPool	N/A	0	0
8	/10/Flatten	Flatten	N/A	0	0
9	/11/Gemm	Gemm	FLOAT	3600	14400
10	/12/Relu	Relu	N/A	0	0
11	/14/Gemm	Gemm	FLOAT	272	1088
12	/15/Relu	Relu	N/A	0	0
13	/16/Gemm	Gemm	FLOAT	68	272
14	out_layer	Output	FLOAT	0	0

Last update: 2026-11-10