A 256-bit vector database implementation in MinZ, demonstrating modern AI/ML algorithms on Z80 hardware.
ZVDB (Z80 Vector Database) is a 1-bit quantized vector similarity search system designed for vintage computing hardware. This MinZ implementation showcases how modern vector database concepts can run efficiently on 1980s processors.
- 256-bit binary vectors for compact representation
- Hamming distance for fast similarity computation
- Popcount optimization with lookup tables (3.3x speedup*)
- K-nearest neighbors search
- Bit-packed structures for 67% memory savings on metadata**
- Comprehensive test suite with 20+ test cases
- Complete SHA256 implementation in pure Z80 assembly (commit 2b870c2)
- High-performance 32-bit operations: ADD32 (34 T-states), XOR32 (56 T-states)
- Page-aligned memory layout for optimal Z80 performance
- ~45,000 T-states per 512-bit block (~11ms @ 4MHz, ~45KB/sec throughput)
- Memory efficient: <2KB total including constants and workspace
- Vector fingerprinting for deduplication and integrity verification
* For bit counting operations specifically
** Metadata only; vectors still require full 256 bits
zvdb-minz/
├── README.md # This file
├── zvdb.minz # Main ZVDB implementation
├── zvdb.a80 # Generated Z80 assembly (for verification)
├── zvdb_test.minz # Test suite
├── zvdb_test.a80 # Test suite assembly (for verification)
├── run_zvdb_tests.sh # Test runner
├── ZVDB_README.md # Performance analysis
├── sha256_z80.asm # 🔥 High-performance SHA256 implementation
├── SHA256_README.md # SHA256 documentation
├── sha256_integration.minz # MinZ/assembly integration examples
├── simple_sha256_demo.minz # Simple MinZ demonstration
└── zvdb_experiments/ # Development iterations
The .a80 assembly files are included so you can:
- Count actual T-states for operations
- Verify the SMC optimizations
- Check the popcount LUT implementation
- See exactly what code runs on your Z80
The SHA256 implementation (sha256_z80.asm) includes:
- All 1,438 lines of optimized Z80 assembly code
- Performance annotations with exact T-state counts
- Test function with known hash verification
- Complete technical documentation
- MinZ compiler v0.9.0+
- Z80 assembler (sjasmplus) for final binary generation
- Z80 hardware or emulator for execution
# Clone the repository
git clone https://github.com/oisee/zvdb-minz
cd zvdb-minz
# Run tests (requires MinZ compiler)
./run_zvdb_tests.sh
# Compile to Z80 assembly
minzc zvdb.minz -o zvdb.a80
# Assemble to binary (requires sjasmplus)
sjasmplus zvdb.a80
# Load at 0x8000 on your Z80 system| Metric | Value | Notes |
|---|---|---|
| Vector size | 256 bits | 32 bytes per vector |
| Popcount speedup | 3.3x | LUT vs bit loop |
| Metadata compression | 67% | Using bit fields |
| Development time | Hours | vs days in assembly |
| Test coverage | 20+ cases | Comprehensive suite |
| Metric | Value | Notes |
|---|---|---|
| Throughput | ~45KB/sec | @ 4MHz Z80 |
| T-states per block | ~45,000 | 512-bit blocks |
| Memory footprint | <2KB | Including workspace |
| ADD32 performance | 34 T-states | Fastest possible on Z80 |
| XOR32 performance | 56 T-states | Optimized via A register |
See ZVDB_README.md for detailed performance analysis with honest benchmarks and disclaimers.
- MinZ Language - The systems programming language for Z80
- ZVDB ABAP Original - Pure ABAP implementation
- ZVDB Z80 ASM Reimplementation - Pure assembly implementation
This is a demonstration project showing MinZ capabilities. Feel free to:
- Port to other vintage platforms
- Optimize further
- Add new similarity metrics
- Improve test coverage
MIT License - See LICENSE file for details
- The MinZ compiler team (=myself and my imaginary friend Claude Code) for making Z80 development fun again
- The Z80 community for keeping vintage computing alive
- Researchers working on 1-bit quantization making AI accessible to all hardware
This implementation has been tested in compilation but not on actual Z80 hardware. Performance claims are based on T-state analysis and estimates. Your mileage may vary. But hey, it compiles! 🎉
Built with ❤️ for the intersection of vintage computing and modern AI