Copper Bars / Plasma

Copper Bars and Plasma

Two related effects from Twisted Brain (Parts 6 and 7), built on the same single-rasterline-rupture chassis: a pre-rendered MODE 0 dither buffer + per-row R12/R13 selection + per-row palette change. The two effects differ only in what’s in the pre-rendered buffer and how it’s indexed.

Common chassis

CRTC config: 64 cycles × 1 character row × 4 scanlines = 256 visible scanlines (see single-rasterline-rupture).

R9 = 3   (4 scanlines per char row)
R4 = 0   (1 char row per cycle)
R6 = 1   (display that row)
R7 = &FF (no VSync this cycle)
R12/R13 → varies per cycle (picked from lookup)

Final 64th cycle restores VSync: R4 = 14 (15 char rows), R7 = 7 for VSync at line 280 (= (280-252)/4).

Pre-rendered buffer: 17 dither patterns from a 4×4 ordered (Bayer) dithering matrix, each pattern stored as one MODE 0 character row. Patterns go pure-black → 4×4-Bayer-fades → pure-white. With ordered dithering the patterns are static, so motion doesn’t scintillate.

Copper Bars (Part 6)

Each cycle picks a screen address pointing at one of the 17 dither rows. Stepping through the table white → black → white → black gives copper-style horizontal bands.

.loop                       ; 64 iterations, must total 512c each
    LDA addr_table_HI, X
    STA &FE00 = 12          ; via prior LDA #12 / STA
    ; ... R12 + R13 writes
    LDA palette_table, X
    STA &FE21               ; palette change in horizontal blank
    ; ... pad to 512c
    INX
    BNE loop

Three things layered on the per-row R12/R13 selection:

1. Hue rotation — palette colour 1 (the dither’s “white” lane) is cycled through the colour wheel: red → magenta → blue → cyan → green → yellow → red. Because MODE 0 only has colours 0 and 1, this is 8 palette writes per change. Writes are scheduled for horizontal blank periods so partial-palette frames aren’t visible. Only update palette when a solid colour is on screen — never mid-dither — so a partially-programmed palette doesn’t bleed across pixels.

2. Scroll — index into the screen-address lookup is incremented per frame so bars scroll up.

3. Sine-stretch — an accumulator adds a per-row delta to the lookup index. Small delta → bars stretched (slow advance through the table); large delta → squashed. The delta is itself a sine wave per frame, so the bars zoom in and out as they scroll.

Plasma (Part 7)

Same CRTC config, same palette tricks, different pre-rendered buffer: instead of 17 black-to-white dither levels, the buffer contains gradient rows of increasing frequency — 1 white-black-white cycle in the top rows, then 2, then 4, then 8 in the bottom rows.

The per-cycle screen-address lookup picks rows by frequency, and each row can be shifted horizontally by adjusting the low bits of R12/R13 (offsetting within the pre-rendered char row).

Animation:

• Shift all rows by the same delta → horizontal scroll.
• Apply a sine wave to per-row shifts → “bend” the bars.
• Add a second sine wave at a different frequency → interfering ripples.
• Vary phase rate per character row → vertical motion of the ripples.

Two colours only, chosen to be close in hue (red/magenta, white/yellow) so the dither blends cleanly to the eye.

No multiplication needed at runtime — everything is sine-table lookup + ADC. Just be careful: it’s easy to produce static-looking patterns; pleasing motion is a black art of phase/frequency tweaking.

Why it works

The trick is the separation of “what content” (pre-rendered dither in screen RAM) from “where on screen each chunk lands” (per-cycle R12/R13). Once you accept that the screen-RAM-to-display mapping is yours to choose, the 2 MHz CPU has plenty of time to update which rows are picked and which palette they use — without ever redrawing a pixel.

A 256-line MODE 0 screen has 20 480 bytes. Filling that at 2 MHz costs 10 240 cycles of pure writes — about 80 raster lines just to clear, never mind drawing. Per-row R12/R13 reuse means we display all 64 character rows from just a 1 KB pre-rendered buffer, and the per-frame work is < 1 KB of CRTC + palette register writes.

Constant-time discipline

Each of the 64 cycles must consume exactly 4 raster lines = 512 cycles. The per-iteration work (compute address, write R12/R13, optionally write palette) takes ~50-100c; the rest is NOP padding or BIT 0 (3c) fillers. See single-rasterline-rupture for the general patterns.

Variants

• Different dither matrix — Bayer 4×4 is the obvious choice; 8×8 would give finer gradients but cost a larger pre-rendered buffer.
• MODE 1 or MODE 2 — would give more palette colours per buffer entry, but bigger pre-rendered buffer and you’d need to be cleverer about which colours to dither. The MODE 0 + hue-rotation combo is the cheapest path to “copper” look.
• Text mode with palette-only effect — see twisted-brain Part 3 (Text Screens). Per-raster palette swap without any vertical rupture, MODE 1. Cheaper but limited to colour effects only.

Builds on / used by

• single-rasterline-rupture — the 4-scanlines-per-cycle chassis.
• vertical-rupture — the general technique.
• video-ula — palette register layout, MODE 0 colour encoding.
• crtc-6845 — R12/R13 latch timing.

---

This wiki is curated by Claude following the LLM-Wiki methodology — a human curates source documents, the LLM compiles structured cross-linked markdown. Content may contain errors, omissions, or stale claims. For authoritative information refer to the original source documents in the bbc-documents GitHub archive.