Improve LinearSolveAutotune UI/UX

[ChrisRackauckas-Claude]

Summary

This PR significantly improves the UI/UX of LinearSolveAutotune by:

• Removing authentication/token prompts from the main autotune flow
• Adding progress bars with real-time status updates
• Creating an AutotuneResults object with beautiful display output
• Separating benchmarking from result sharing
• Replacing the binary large_matrices flag with flexible size categories
• Adding clear documentation for GitHub CLI setup

Key Changes

1. Progress Bar and Status Updates

• Added ProgressMeter to show real-time progress during benchmarking
• Shows current algorithm being tested, matrix size, and element type
• Percentage completion visible throughout the process

2. AutotuneResults Object

• New AutotuneResults struct that wraps results, system info, and plots
• Beautiful display output showing:
  • System information summary
  • Top performing algorithms with GFLOPs
  • Element types and matrix sizes tested
  • Clear call-to-action for sharing results
• plot(results) creates composite plots of all benchmarks
• share_results(results) for easy community contribution

3. Independent Autotune and Sharing

• autotune_setup() now runs benchmarks without any authentication
• New share_results() function handles authentication and telemetry separately
• Users can benchmark privately and choose to share results later

4. Flexible Matrix Size Categories

Replaced large_matrices::Bool with sizes::Vector{Symbol}:

• :small - 5×5 to 20×20 matrices
• :medium - 20×20 to 300×300 matrices (expanded range)
• :large - 300×300 to 1000×1000 matrices
• :big - 10000×10000 to 100000×100000 matrices

Default now includes :large for better coverage.

5. Simplified Authentication

• Removed interactive token input prompts
• Clear instructions for setting up gh CLI
• Falls back to environment variable if available
• No more interrupting the benchmark flow

6. Improved Documentation

• Added comprehensive README with setup instructions
• Clear guidance on GitHub CLI installation and authentication
• Examples for all common use cases

Example Usage

# Run benchmarks with new defaults (includes large matrices)
results = autotune_setup()

# Beautiful display output
display(results)
# Shows:
# - System info
# - Top algorithms ranked by performance
# - Clear instructions to share results

# Create composite plots
plot(results)

# Share with community (optional)
share_results(results)

Breaking Changes

• large_matrices parameter replaced with sizes
• telemetry parameter removed from autotune_setup()
• Returns AutotuneResults object instead of tuple
• Result sharing now requires explicit share_results() call

Migration Guide

Old:

results_df, sysinfo, plots = autotune_setup(large_matrices = true, telemetry = true)
share_results(results_df, sysinfo, plots)

New:

results = autotune_setup(sizes = [:small, :medium, :large, :big])
display(results)  # See beautiful summary
plot(results)     # Create plots
share_results(results)  # Share with community

Testing

• Package loads successfully
• Syntax checks pass
• Progress bar displays correctly
• AutotuneResults display formatting works

🤖 Generated with Claude Code

[ChrisRackauckas]

Suggested change

	@info " You can manually share this file on the issue tracker."
	@info " You can manually share this file on the issue tracker:"
	@info " https://github.com/SciML/LinearSolve.jl/issues/669"