Add ck-rocprof: GPU profiling tool for rocprof-compute

[tenpercent]

Summary

Adds a new command-line profiling tool (ck-rocprof) to simplify GPU performance analysis workflow using AMD rocprof-compute.

Key Features

• Easy setup - Automatic Python venv configuration with one command
• Simple CLI - Intuitive commands: setup, run, analyze, compare, list
• Auto GPU detection - Automatically detects GPU architecture (e.g., gfx950, MI350)
• LDS metrics focus - Specializes in Local Data Share (Block 12) metrics for bank conflict analysis
• Comprehensive docs - Detailed documentation with examples and troubleshooting

Usage Examples

# One-time setup
ck-rocprof setup

# Profile an executable
ck-rocprof run baseline ./bin/tile_example_gemm_universal

# Analyze LDS metrics
ck-rocprof analyze baseline

# Compare two profiling runs
ck-rocprof compare baseline optimized

# List available runs
ck-rocprof list

Files Added

• script/tools/ck-rocprof (414 lines) - Main profiling tool script
• script/tools/ck-rocprof.md (415 lines) - Complete documentation

Benefits

• Streamlines GPU profiling workflow for CK developers
• Reduces complexity of using rocprof-compute directly
• Enables easier performance optimization and bank conflict detection
• Provides convenient comparison tools for A/B testing kernel changes

Test Plan

• Manual testing with various CK executables
• Verified setup process creates proper Python venv
• Validated profiling runs capture LDS metrics correctly
• Tested comparison functionality between multiple runs
• Confirmed documentation examples work as expected