Add roofline_analyzer 0.1.1

[tykow]

I've created a python cli application that
as an input takes the XLSX that is, to my knowledge. generated by generate_perf_report.py from
#138

and when combined with a config-file that has e.g. MI300X bfloat16 values:

It will generate a naive roofline analysis for the GEMMs used in e.g. a client model:

and it will generate a new XLSX as an output, that has:

1. bound_type: what is the first limiting roofline (memory or compute), based on the "AI ridge point"
2. reference_roofline: what is used as the reference value for roofline distance calculation
3. bound_distance: what is the distance from kernel to the maf_roofline in question
4. bound_distance_pct: above in percentage

and AI ridge points for both the max achievable flops (maf) and max theoretical flops (mtf) i.e. the point that divides whether the kernel is memory or compute limited:

What is new in 0.1.1 compared to 0.1.0 is that the kernels will get a rolling numbering, which is also used to indicate which point on the plot belongs to which kernel:

For additional information, consult the extensive README.md

[Copilot]

Pull Request Overview

Adds version 0.1.1 of the roofline_analyzer CLI tool, introducing kernel ID numbering for plot annotations and full end-to-end Excel-based roofline analysis.

• Introduces an ID column for kernels and ties IDs into plot annotations.
• Implements modular functions for reading, analyzing, plotting, and exporting Excel data.
• Updates project metadata (pyproject.toml), sample config, and documentation for 0.1.1.

Reviewed Changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
examples/custom_workflows/roofline_analyzer/src/xlsx.py	Excel data parsing, analysis column addition, and export functions
examples/custom_workflows/roofline_analyzer/src/roofline.py	Roofline model calculations and plot generation
examples/custom_workflows/roofline_analyzer/src/cli.py	Command-line interface for config handling and summary display
examples/custom_workflows/roofline_analyzer/pyproject.toml	Project metadata, version bump to 0.1.1, dependencies, and entry point
examples/custom_workflows/roofline_analyzer/mi300x_bfloat16.toml	Sample accelerator configuration for MI300X bfloat16
examples/custom_workflows/roofline_analyzer/main.py	Main CLI command integrating all modules
examples/custom_workflows/roofline_analyzer/README.md	Installation, usage instructions, and output schema documentation
examples/custom_workflows/roofline_analyzer/.gitignore	Ignore patterns for build artifacts and generated files

Comments suppressed due to low confidence (2)

examples/custom_workflows/roofline_analyzer/src/xlsx.py:1

• Key analysis and export functions lack accompanying unit tests; consider adding tests for read_xlsx, add_analysis_columns, and export_to_xlsx to verify behavior and guard against regressions.

import pandas as pd

examples/custom_workflows/roofline_analyzer/README.md:93

• [nitpick] Markdown bullets for bound_distance and bound_distance_pct aren't indented under the nested list for the analyzed sheet section; adjust indentation for consistent formatting.

* **bound_distance**: Distance to the nearest roofline

[Copilot]

The parameter ai_ridge_mtf is never used in add_analysis_columns; consider removing it or using it as intended to keep the signature accurate.

Suggested change

ai_ridge_mtf: float,

[Copilot]

Mutating the passed config dict inside this function may lead to unexpected side effects; use a local variable for kernel_name_column instead of modifying the original config.

Suggested change

	# Set config to use 'ID' as kernel_name_column for plotting
	config['excel']['kernel_name_column'] = 'ID'
	# Use a local variable for 'ID' as kernel_name_column for plotting
	kernel_name_column = 'ID'