Warning

This is vibe coded slop. You should never use it unless you are truly brave.

GPU-accelerated Unix utilities with Zig frontends and Metal/Vulkan compute backends.

Each utility is available in two variants:

# Pure builds (self-contained, no dependencies)
brew tap e-jerk/utils
brew install e-jerk/utils/utils          # all 4 tools
brew install e-jerk/utils/grep           # individual

# GNU builds (--gnu flag for full POSIX compliance)
brew tap e-jerk/utils-gnu
brew install e-jerk/utils-gnu/utils-gnu  # all 4 tools
brew install e-jerk/utils-gnu/sed        # individual

macOS: Add Homebrew bin to PATH to override Apple's built-in tools:

echo 'export PATH="$(brew --prefix)/bin:$PATH"' >> ~/.zshrc

All utilities support:

• --auto — Automatically select optimal backend (default)
• --gpu — Force GPU (Metal on macOS, Vulkan on Linux)
• --cpu — Force CPU backend (SIMD-optimized)
• --gnu — Force GNU implementation (gnu build only)
• --metal — Force Metal backend (macOS only)
• --vulkan — Force Vulkan backend

Use -V or --verbose to see timing and backend information.

Total: 147 GNU compatibility tests passing

Each tool includes native implementations with GPU and SIMD optimization:

Optimization key: [GPU+SIMD] means the feature runs on GPU compute shaders (Metal/Vulkan) with SIMD-optimized CPU fallback. [GPU VM] means features are compiled to bytecode and executed on a GPU virtual machine. Use --help on any tool to see optimization annotations for each flag.

Use --gnu flag (gnu build only) for unsupported features to fall back to GNU implementation.

All tools share the e_jerk_gpu library for:

• Hardware capability detection (threads, memory, buffer limits)
• Performance-based backend scoring (Entry/Mid/High/Ultra tiers)
• Automatic CPU/GPU selection based on workload characteristics
• Swappable compute backend interface

• CPU: Zig @Vector(16, u8) and @Vector(32, u8) for parallel byte operations
• Metal: uchar4 vectorized operations (4-byte SIMD)
• Vulkan: uvec4 vectorized operations (16-byte SIMD)

Results measured on Apple M1 Max.

Each tool uses the same build pattern:

# Build with optimizations
zig build -Doptimize=ReleaseFast

# Run tests
zig build test      # Unit tests
zig build smoke     # Integration tests (GPU verification)
zig build bench     # Benchmarks
bash gnu-tests.sh   # GNU compatibility tests

Requirements:

• Zig 0.15.2+
• glslc (for Vulkan shader compilation)
• macOS Metal or Vulkan runtime

• Native -A/-B/-C context lines with proper group separators
• Native -r recursive search with combined options
• Native --color output with ANSI highlighting

• Native -e multiple expressions
• Native line addressing (2s/..., 2,4s/..., $s/...)

• Native -size [+-]N[ckMG] file size filtering
• Native -mtime/-atime/-ctime [+-]N time filtering
• Native -prune PATTERN directory pruning

• GPU Bytecode VM: Complex AWK programs compile to bytecode and execute on GPU
  • Supports: variables, arithmetic, comparisons, control flow (if/while/for), math functions
  • Each GPU thread processes one input line independently
• Native built-in functions: length(), substr(), index(), toupper(), tolower(), sin(), cos(), sqrt(), log(), exp(), int()
• Native special variables: NR, NF
• Native GPU regex: Thompson NFA execution on Metal and Vulkan
• Full backend parity (CPU, Metal, Vulkan produce identical results)

Source code: Unlicense (public domain) Binaries: GPL-3.0-or-later (due to linked dependencies)