Developer ports a software 3D points renderer to ZX Spectrum 48K in Z80 assembly, pushing from 6.2 fps in C to 14 fps in ASM and 40 fps with full precomputation.
Key Takeaways
Toolchain: z88dk cross-compiler with a simple Makefile; outputs .tap files runnable in the FUSE emulator; repo includes prebuilt taps for statue and sphere models.
Core optimization: replaced two runtime divisions with reciprocal lookup table multiplications and used Z80 page-based (HL) register tricks to avoid costly memory addressing.
Precompute branch encodes pixel VRAM addresses and bit offsets into 16 bits/pixel at build time, leaving the inner loop as near-pure reads and writes; 4x speedup over runtime ASM.
Inline Z80 ASM blitter runs 3.5x faster than equivalent C for the same reads/shifts/writes, confirming Z80 C compilers leave significant register efficiency on the table.
Projection math uses no runtime multiplications: two divisions plus additions, derived from standard perspective equations with a fixed orbit camera and pre-scaled integer coordinates (S=8960).
Hacker News Comment Review
Commenters converge on tooling as the key insight: modern assembler IDEs and cross-compilers make 8-bit assembly far more productive than 1980s workflows, narrowing the gap with high-level languages.
A second commenter is independently porting modern renderers to the Spectrum and flagged the reciprocal lookup and axis-swap tricks as ideas worth adopting.
