2217 add io module for profiler

[smedegaard]

Proposed changes

As described in the issue this is an effort to extend and improve the output of CK Profiler.

Discussion

This pull request is more to have a conversation around than a final implementation.

Disclaimer, the author is not experienced in C++. Input and feedback on how to do things in a more idiomatic way is welcome.

What's in the PR

This PR includes

• adding argparse for easier parsing of arguments
• io_profiler
  • support for output to:
    • console (the default option. will use the same format as previously)
    • newline delimited json (jsonl)

Output Format

The output format is intended to capture the same data as the current implementations, and be flexible enough to support the different kinds of operations available.

What is NOT in the PR

The PR has only updated profiler/include/profiler/profile_gemm_impl.hpp to use the new io_profiler functions.
If this approach to output is accepted, all the output logic will live in io_profiler. Therefore all the implementation headers must be updated to leverage it.

It does not capture meta data like GPU name, architecture, CK version, etc

Example

Tested on Radeon RX 7900 XT

./bin/ckProfiler gemm 1 2 1 2 0 1 3840 4096 4096 4096 4096 4096 5 10 -o jsonl=result.jsonl

cat result.jsonl

{"operation":"DeviceGemmDl<256, 128, 128, 16, 2, 4, 4, 1>","time_ms":3.20498,"tflops":40.2027,"gb_per_sec":30.0997,"is_best":false,"timestamp":"2025-06-04T11:12:33.466Z","N":4096,"M":3840,"layout_b":"RowMajor","output_datatype":"unknown","K":4096,"weight_datatype":"unknown","input_datatype":"unknown","layout_c":"RowMajor","layout_a":"ColumnMajor","datatype":"unknown","operation_type":"gemm","operation_params":{"N":4096,"M":3840,"layout_b":"RowMajor","output_datatype":"unknown","K":4096,"weight_datatype":"unknown","input_datatype":"unknown","layout_c":"RowMajor","layout_a":"ColumnMajor","datatype":"unknown","operation_type":"gemm"}}
{"operation":"DeviceGemmDl<256, 128, 128, 16, 2, 4, 4, 1>","time_ms":3.14917,"tflops":40.9153,"gb_per_sec":30.6332,"is_best":false,"timestamp":"2025-06-04T11:12:33.655Z","N":4096,"M":3840,"layout_b":"RowMajor","output_datatype":"unknown","K":4096,"weight_datatype":"unknown","input_datatype":"unknown","layout_c":"RowMajor","layout_a":"ColumnMajor","datatype":"unknown","operation_type":"gemm","operation_params":{"N":4096,"M":3840,"layout_b":"RowMajor","output_datatype":"unknown","K":4096,"weight_datatype":"unknown","input_datatype":"unknown","layout_c":"RowMajor","layout_a":"ColumnMajor","datatype":"unknown","operation_type":"gemm"}}
...

Output truncated for brevity