dspk benchmark

Time processing on the selected runtime

dspk benchmark [OPTIONS] <COMMAND>

dspk benchmark pipeline

A PipelineWorkspace (scale → notch → TKEO): upload, stages, download

dspk benchmark pipeline [OPTIONS]

Options

• -c, --channels <CHANNELS>: default 384
  Channels
• -s, --samples <SAMPLES>: default 30000
  Samples per channel
• -r, --sample-rate <SAMPLE_RATE>: default 30000
  Sample rate (Hz): sets the filters and the real-time factor
• -i, --iterations <ITERATIONS>: default 5
  Timed repetitions (the median is reported)
• --int16
  Upload int16 stored samples (scaled on the device) instead of f32
• --save <SAVE>
  Write the output once as raw binary + sidecar (timed separately)

dspk benchmark sweep

The pipeline’s stages across channel counts

dspk benchmark sweep [OPTIONS]

Options

• -c, --channels <CHANNELS>: default 384
  Channels
• -s, --samples <SAMPLES>: default 30000
  Samples per channel
• -r, --sample-rate <SAMPLE_RATE>: default 30000
  Sample rate (Hz): sets the filters and the real-time factor
• -i, --iterations <ITERATIONS>: default 5
  Timed repetitions (the median is reported)

dspk benchmark suite

Every kernel family and streaming detection; writes a JSON report

dspk benchmark suite [OPTIONS] --report-dir <REPORT_DIR>

Options

• -c, --channels <CHANNELS>: default 384
  Channels
• -s, --samples <SAMPLES>: default 30000
  Samples per channel
• -r, --sample-rate <SAMPLE_RATE>: default 30000
  Sample rate (Hz): sets the filters and the real-time factor
• -i, --iterations <ITERATIONS>: default 5
  Timed repetitions (the median is reported)
• --report-dir <REPORT_DIR>
  Folder for the report (<unix time>-<runtime>.json)