Parallax Bars (64-row buffer via mid-frame ACCCON switch)

Parallax Bars

Vertical parallax-scrolling bars in MODE 1, inspired by Rebels’ Total Triple Trouble on the Amiga. Built on single-rasterline-rupture with one extra ingredient: a 64-character-row pre-rendered buffer split across main RAM and SHADOW RAM, switched mid-frame via ACCCON (&FE34) — see shadow-ram.

From Twisted Brain Part 8 (twisted-brain).

The problem

7 layers of bars at decreasing distance from a virtual “camera”. The top layer is 32 px wide × 32 px apart = 5 bars across a 320-pixel MODE 1 screen. Move the camera 1 pixel right and the bars all shift by amounts proportional to their distance. After 64 single-pixel camera-shifts the front layer wraps and we’re back where we started.

So we need 64 character rows of pre-rendered bars (each row is one camera-position snapshot). The CRTC then picks the right row for each scanline and the user sees parallax motion as the row choice scrolls.

Why two screens

MODE 1 screen = 32 character rows × 80 bytes = 20 KB. Our 64-row buffer needs 40 KB — too big for one screen.

Solution: split the buffer across two MODE 1 screens, one in main RAM (rows 0-31) and one in SHADOW RAM (rows 32-63). Switch which the CRTC reads from per cycle.

The mid-frame switch

CRTC R12/R13 is latched at the start of each cycle (see crtc-6845). ACCCON D bit (display source) takes effect immediately. So the sequence per cycle is:

1. During this cycle: write the next cycle’s R12/R13 (latched).
2. Immediately before the next cycle starts: write ACCCON to switch main↔SHADOW.

Timing has to be sharp — ACCCON is read by the address translator on the very next CRTC fetch. If you write it too early, the current cycle’s display flips mid-fetch. If you write it too late, the new cycle starts with the wrong RAM source.

The window: the end of the 4th scanline (horizontal blank) of the current cycle. By then all displayable bytes for the current cycle have been fetched, and the next cycle’s first fetch hasn’t happened yet.

The per-cycle loop

CRTC config: 64 cycles per frame = 63 four-scanline display cycles + 1 final rebalance cycle (same chassis as copper-bars — see single-rasterline-rupture for the accounting). The draw function enters with cycle 1 already underway and the loop iterates 62 times to set up cycles 2 through 63.

Per-iteration work (must fit in 4 raster lines = 512c):

.loop                       ; 62 iterations
    ; --- per-frame sine math ---
    TXA                     ; 2c
    CLC                     ; 2c
    ADC parallax_wavey      ; 3c
    TAX                     ; 2c
    LDA sine_table, X       ; 4c
    CLC                     ; 2c
    ADC parallax_x          ; 3c
    AND #&3F                ; 2c   wrap to 0..63
    TAY                     ; 2c   Y indexes the buffer-row table
 
    ; pad until horizontal blank
    FOR n,1,23 : NOP : NEXT : BIT 0   ; 49c
 
    JSR cycles_wait_128     ; +128c   wait one raster line
    JSR cycles_wait_128     ; +128c   wait another
 
    ; --- write R12/R13 for next cycle ---
    LDA #12 : STA &FE00 : LDA vram_hi, Y : STA &FE01   ; 18c
    LDA #13 : STA &FE00 : LDA vram_lo, Y : STA &FE01   ; 18c
 
    JSR cycles_wait_128     ; +128c   one more raster line
 
    ; --- write ACCCON for next cycle (in 4th hsync) ---
    LDA &FE34               ; 4c (stretched)
    AND #&FE                ; 2c   clear D bit
    ORA vram_page, Y        ; 4c   set D bit if row >= 32
    STA &FE34               ; 4c (stretched)
 
    DEC parallax_crtc_row   ; 5c
    BNE loop                ; 3c

The three JSR cycles_wait_128 calls span the three “passive” raster lines per cycle. The fourth raster line ends in the ACCCON write.

Animation

• Per character row: index into the buffer is (parallax_x + sine_table[parallax_wavey + row]) & 63. So bars at different rows are offset by a sine wave — gives the wavy parallax-with-curvature look.
• Per frame: increment parallax_x (horizontal scroll) and parallax_wavey (phase of the wave).
• All offsets are sine-table lookup + ADC — no runtime multiplication.

Why it’s the most timing-sensitive effect in Twisted Brain

If Timer 1 fires even one raster line early or late, the ACCCON write lands one raster line off, and the main↔SHADOW switch happens during the wrong cycle. Result: single-pixel “glitch” rows at the main/SHADOW boundary. Particularly visible at the row 31→32 transition (where the buffer source actually changes).

This is the canonical failure mode that exposes the Kefrens-bars 313-line frame bug — see kefrens-bars for why that bug propagates here and how the kill function uses a 311-line rebalance frame to put Timer 1 back where it should be.

The pre-rendered buffer

Each of the 64 rows is one “camera offset” with all 7 bar layers drawn. Generated with a BASIC program that:

1. Draws each layer from back to front at the appropriate distance-scaled offset.
2. Uses ordered dither inside each bar to give a smooth gradient, with the dither pattern keyed to pixel coordinates within the bar (not screen coordinates) — so the dither stays attached to the bar and doesn’t scintillate as bars move.

40 KB of pre-rendered art is a lot of memory, but it’s only paid once at init time.

Variants worth knowing

• MODE 0 version: would let you have 80-pixel-wide bars with finer dither, but doubles the per-row buffer size — would need 80 KB pre-rendered. Doesn’t fit.
• Procedural bars at runtime: in principle possible if you can compute one row’s pixels in <512c. With 80 bytes to write per row, you’d have ~6 cycles/byte budget — feasible for plain colour bars, but no dithering.
• More layers: each additional layer doubles the camera-offset cycle, so 8 layers would need 128 buffer rows / 80 KB. Not viable in 64 KB main + 64 KB SHADOW.

Builds on / used by

• single-rasterline-rupture — the 4-scanlines-per-cycle chassis (same as copper-bars).
• shadow-ram — ACCCON bit layout, D vs E vs X bits.
• crtc-6845 — R12/R13 latch timing relative to ACCCON immediate effect.
• fx-framework — Timer 1 stable raster (the parallax effect would be impossible without it).

---

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.