Tools

Contents

Tools#

2024-09-30

3 min read time

Applies to Linux

roctx.py#

You can use the roctx command with rocprof binary to get marker timing information for each MIGraphX operator. To process timing information, use roctx.py helper script.

Usage: roctx.py [-h] [--json-path json_path] [--out out]
[--study-name study-name] [--repeat repeat] [--parse]
[--run run] [--debug]

The roctx.py helper script provides two main functionalities: run and parse.

--run#

Runs migraphx-driver roctx command with the given migraphx-driver knobs followed by the parsing of the result which provides GPU kernel timing information. You can pass the MIGraphX knobs via a string to –run knob. See the _roctx-examples for usage.

--parse#

Parses JSON file in the given --json-path and provides GPU kernel timing information.

--out#

Output folder

--study-name#

Optional. Allows user to name a study for easy interpretation. Defaults to timestamp.

--repeat#

Number of iterations. Sets to 2 by default.

--debug#

Provides additional debug information related to data. Use for debugging purposes only.

Examples:

Running inference with rocTX for a given ONNX file:

python roctx.py --run '--onnx --gpu fcn-resnet50-11.onnx' --out output_folder --repeat 5

Example output:

../_images/roctx1.jpg

Hotspot kerel timing information:

../_images/roctx2.jpg

The output provides SUM, MIN, MAX and COUNT information for each kernel executed for a given model. It also provides the average total time. The following three files are provided for reference:

  • OUTPUT CSV FILE: Provides a summary of the run which includes utilized MIGraphX knobs and related kernel timing information.

  • KERNEL TIMING DETAILS: Provides the hotspot kernel timing information.

  • ALL DATA FROM ALL RUNS: Provides all output data related to all iterations executed during a run.

Parsing an existing JSON file:

python roctx.py --parse --json-path ../trace.json
Contents